Sage Source Browser

schemes/elliptic_curves/padics.py (browse directory)

# -*- coding: utf-8 -*- """ Miscellaneous p-adic functions p-adic functions from ell_rational_field.py, moved here to reduce crowding in that file. """ ###################################################################### # Copyright (C) 2007 William Stein <wstein@gmail.com> # # Distributed under the terms of the GNU General Public License (GPL) # # This code is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU # General Public License for more details. # # The full text of the GPL is available at: # # http://www.gnu.org/licenses/ ###################################################################### import sage.rings.all as rings import padic_lseries as plseries import sage.rings.arith as arith from sage.rings.all import ( Qp, Zp, Integers, Integer, O, PowerSeriesRing, LaurentSeriesRing, RationalField) import math import sage.misc.misc as misc import sage.matrix.all as matrix sqrt = math.sqrt import monsky_washnitzer import sage.schemes.hyperelliptic_curves.hypellfrob def __check_padic_hypotheses(self, p): r""" Helper function that determines if `p` is an odd prime of good ordinary reduction. EXAMPLES:: sage: E = EllipticCurve('11a1') sage: from sage.schemes.elliptic_curves.padics import __check_padic_hypotheses sage: __check_padic_hypotheses(E,5) 5 sage: __check_padic_hypotheses(E,29) Traceback (most recent call last): ... ArithmeticError: p must be a good ordinary prime """ p = rings.Integer(p) if not p.is_prime(): raise ValueError, "p = (%s) must be prime"%p if p == 2: raise ValueError, "p must be odd" if self.conductor() % p == 0 or self.ap(p) % p == 0: raise ArithmeticError, "p must be a good ordinary prime" return p def padic_lseries(self, p, normalize='L_ratio', use_eclib=False): r""" Return the `p`-adic `L`-series of self at `p`, which is an object whose approx method computes approximation to the true `p`-adic `L`-series to any desired precision. INPUT: - ``p`` - prime - ``use_eclib`` - bool (default:False); whether or not to use John Cremona's eclib for the computation of modular symbols - ``normalize`` - 'L_ratio' (default), 'period' or 'none'; this is describes the way the modular symbols are normalized. See modular_symbol for more details. EXAMPLES:: sage: E = EllipticCurve('37a') sage: L = E.padic_lseries(5); L 5-adic L-series of Elliptic Curve defined by y^2 + y = x^3 - x over Rational Field sage: type(L) <class 'sage.schemes.elliptic_curves.padic_lseries.pAdicLseriesOrdinary'> We compute the `3`-adic `L`-series of two curves of rank `0` and in each case verify the interpolation property for their leading coefficient (i.e., value at 0):: sage: e = EllipticCurve('11a') sage: ms = e.modular_symbol() sage: [ms(1/11), ms(1/3), ms(0), ms(oo)] [0, -3/10, 1/5, 0] sage: ms(0) 1/5 sage: L = e.padic_lseries(3) sage: P = L.series(5) sage: P(0) 2 + 3 + 3^2 + 2*3^3 + 2*3^5 + 3^6 + O(3^7) sage: alpha = L.alpha(9); alpha 2 + 3^2 + 2*3^3 + 2*3^4 + 2*3^6 + 3^8 + O(3^9) sage: R.<x> = QQ[] sage: f = x^2 - e.ap(3)*x + 3 sage: f(alpha) O(3^9) sage: r = e.lseries().L_ratio(); r 1/5 sage: (1 - alpha^(-1))^2 * r 2 + 3 + 3^2 + 2*3^3 + 2*3^5 + 3^6 + 3^7 + O(3^9) sage: P(0) 2 + 3 + 3^2 + 2*3^3 + 2*3^5 + 3^6 + O(3^7) Next consider the curve 37b:: sage: e = EllipticCurve('37b') sage: L = e.padic_lseries(3) sage: P = L.series(5) sage: alpha = L.alpha(9); alpha 1 + 2*3 + 3^2 + 2*3^5 + 2*3^7 + 3^8 + O(3^9) sage: r = e.lseries().L_ratio(); r 1/3 sage: (1 - alpha^(-1))^2 * r 3 + 3^2 + 2*3^4 + 2*3^5 + 2*3^6 + 3^7 + O(3^9) sage: P(0) 3 + 3^2 + 2*3^4 + 2*3^5 + O(3^6) We can use eclib to compute the `L`-series:: sage: e = EllipticCurve('11a') sage: L = e.padic_lseries(3,use_eclib=True) sage: L.series(5,prec=10) # NOTE: Output below prints slightly incorrectly -- see trac 10280! 2 + 3 + 3^2 + 2*3^3 + 2*3^5 + 3^6 + O(3^7) + (1 + 3 + 2*3^2 + 3^3 + O(3^4))*T + (1 + 2*3 + O(3^4))*T^2 + (3 + 2*3^2 + O(3^3))*T^3 + (2*3 + 3^2 + O(3^3))*T^4 + (2 + 2*3 + 2*3^2 + O(3^3))*T^5 + (1 + 3^2 + O(3^3))*T^6 + (2 + 3^2 + O(3^3))*T^7 + (2 + 2*3 + 2*3^2 + O(3^3))*T^8 + (2 + O(3^2))*T^9 + O(T^10) """ key = (p, normalize) try: return self._padic_lseries[key] except AttributeError: self._padic_lseries = {} except KeyError: pass if self.ap(p) % p != 0: Lp = plseries.pAdicLseriesOrdinary(self, p, normalize = normalize, use_eclib=use_eclib) else: Lp = plseries.pAdicLseriesSupersingular(self, p, normalize = normalize, use_eclib=use_eclib) self._padic_lseries[key] = Lp return Lp def padic_regulator(self, p, prec=20, height=None, check_hypotheses=True): r""" Computes the cyclotomic p-adic regulator of this curve. INPUT: - ``p`` - prime = 5 - ``prec`` - answer will be returned modulo `p^{\mathrm{prec}}` - ``height`` - precomputed height function. If not supplied, this function will call padic_height to compute it. - ``check_hypotheses`` - boolean, whether to check that this is a curve for which the p-adic height makes sense OUTPUT: The p-adic cyclotomic regulator of this curve, to the requested precision. If the rank is 0, we output 1. TODO: - remove restriction that curve must be in minimal Weierstrass form. This is currently required for E.gens(). AUTHORS: - Liang Xiao: original implementation at the 2006 MSRI graduate workshop on modular forms - David Harvey (2006-09-13): cleaned up and integrated into Sage, removed some redundant height computations - Chris Wuthrich (2007-05-22): added multiplicative and supersingular cases - David Harvey (2007-09-20): fixed some precision loss that was occurring EXAMPLES:: sage: E = EllipticCurve("37a") sage: E.padic_regulator(5, 10) 5 + 5^2 + 5^3 + 3*5^6 + 4*5^7 + 5^9 + O(5^10) An anomalous case:: sage: E.padic_regulator(53, 10) 26*53^-1 + 30 + 20*53 + 47*53^2 + 10*53^3 + 32*53^4 + 9*53^5 + 22*53^6 + 35*53^7 + 30*53^8 + O(53^9) An anomalous case where the precision drops some:: sage: E = EllipticCurve("5077a") sage: E.padic_regulator(5, 10) 5 + 5^2 + 4*5^3 + 2*5^4 + 2*5^5 + 2*5^6 + 4*5^7 + 2*5^8 + 5^9 + O(5^10) Check that answers agree over a range of precisions:: sage: max_prec = 30 # make sure we get past p^2 # long time sage: full = E.padic_regulator(5, max_prec) # long time sage: for prec in range(1, max_prec): # long time ... assert E.padic_regulator(5, prec) == full # long time A case where the generator belongs to the formal group already (trac #3632):: sage: E = EllipticCurve([37,0]) sage: E.padic_regulator(5,10) 2*5^2 + 2*5^3 + 5^4 + 5^5 + 4*5^6 + 3*5^8 + 4*5^9 + O(5^10) The result is not dependent on the model for the curve:: sage: E = EllipticCurve([0,0,0,0,2^12*17]) sage: Em = E.minimal_model() sage: E.padic_regulator(7) == Em.padic_regulator(7) True Allow a Python int as input:: sage: E = EllipticCurve('37a') sage: E.padic_regulator(int(5)) 5 + 5^2 + 5^3 + 3*5^6 + 4*5^7 + 5^9 + 5^10 + 3*5^11 + 3*5^12 + 5^13 + 4*5^14 + 5^15 + 2*5^16 + 5^17 + 2*5^18 + 4*5^19 + O(5^20) """ p = Integer(p) # this is assumed in code below if check_hypotheses: if not p.is_prime(): raise ValueError, "p = (%s) must be prime"%p if p == 2: raise ValueError, "p must be odd" # todo if self.conductor() % (p**2) == 0: raise ArithmeticError, "p must be a semi-stable prime" if self.conductor() % p == 0: Eq = self.tate_curve(p) reg = Eq.padic_regulator(prec=prec) return reg elif self.ap(p) % p == 0: lp = self.padic_lseries(p) reg = lp.Dp_valued_regulator(prec=prec) return reg else: if self.rank() == 0: return Qp(p,prec)(1) if height is None: height = self.padic_height(p, prec, check_hypotheses=False) d = self.padic_height_pairing_matrix(p=p, prec=prec, height=height, check_hypotheses=False) return d.determinant() def padic_height_pairing_matrix(self, p, prec=20, height=None, check_hypotheses=True): r""" Computes the cyclotomic `p`-adic height pairing matrix of this curve with respect to the basis self.gens() for the Mordell-Weil group for a given odd prime p of good ordinary reduction. INPUT: - ``p`` - prime = 5 - ``prec`` - answer will be returned modulo `p^{\mathrm{prec}}` - ``height`` - precomputed height function. If not supplied, this function will call padic_height to compute it. - ``check_hypotheses`` - boolean, whether to check that this is a curve for which the p-adic height makes sense OUTPUT: The p-adic cyclotomic height pairing matrix of this curve to the given precision. TODO: - remove restriction that curve must be in minimal Weierstrass form. This is currently required for E.gens(). AUTHORS: - David Harvey, Liang Xiao, Robert Bradshaw, Jennifer Balakrishnan: original implementation at the 2006 MSRI graduate workshop on modular forms - David Harvey (2006-09-13): cleaned up and integrated into Sage, removed some redundant height computations EXAMPLES:: sage: E = EllipticCurve("37a") sage: E.padic_height_pairing_matrix(5, 10) [5 + 5^2 + 5^3 + 3*5^6 + 4*5^7 + 5^9 + O(5^10)] A rank two example:: sage: e =EllipticCurve('389a') sage: e._set_gens([e(-1, 1), e(1,0)]) # avoid platform dependent gens sage: e.padic_height_pairing_matrix(5,10) [ 3*5 + 2*5^2 + 5^4 + 5^5 + 5^7 + 4*5^9 + O(5^10) 5 + 4*5^2 + 5^3 + 2*5^4 + 3*5^5 + 4*5^6 + 5^7 + 5^8 + 2*5^9 + O(5^10)] [5 + 4*5^2 + 5^3 + 2*5^4 + 3*5^5 + 4*5^6 + 5^7 + 5^8 + 2*5^9 + O(5^10) 4*5 + 2*5^4 + 3*5^6 + 4*5^7 + 4*5^8 + O(5^10)] An anomalous rank 3 example:: sage: e = EllipticCurve("5077a") sage: e._set_gens([e(-1,3), e(2,0), e(4,6)]) sage: e.padic_height_pairing_matrix(5,4) [4 + 3*5 + 4*5^2 + 4*5^3 + O(5^4) 4 + 4*5^2 + 2*5^3 + O(5^4) 3*5 + 4*5^2 + 5^3 + O(5^4)] [ 4 + 4*5^2 + 2*5^3 + O(5^4) 3 + 4*5 + 3*5^2 + 5^3 + O(5^4) 2 + 4*5 + O(5^4)] [ 3*5 + 4*5^2 + 5^3 + O(5^4) 2 + 4*5 + O(5^4) 1 + 3*5 + 5^2 + 5^3 + O(5^4)] """ if check_hypotheses: p = __check_padic_hypotheses(self, p) K = Qp(p, prec=prec) rank = self.rank() M = matrix.matrix(K, rank, rank, 0) if rank == 0: return M basis = self.gens() if height is None: height = self.padic_height(p, prec, check_hypotheses=False) # Use <P, Q> =1/2*( h(P + Q) - h(P) - h(Q) ) for i in range(rank): M[i,i] = height(basis[i]) for i in range(rank): for j in range(i+1, rank): M[i, j] = ( height(basis[i] + basis[j]) - M[i,i] - M[j,j] ) / 2 M[j, i] = M[i, j] return M def _multiply_point(E, R, P, m): r""" Computes coordinates of a multiple of P with entries in a ring. INPUT: - ``E`` - elliptic curve over Q with integer coefficients - ``P`` - a rational point on P that reduces to a non-singular point at all primes - ``R`` - a ring in which 2 is invertible (typically `\ZZ/L\ZZ` for some positive odd integer `L`). - ``m`` - an integer, = 1 OUTPUT: A triple `(a', b', d')` such that if the point `mP` has coordinates `(a/d^2, b/d^3)`, then we have `a' \equiv a`, `b' \equiv \pm b`, `d' \equiv \pm d` all in `R` (i.e. modulo `L`). Note the ambiguity of signs for `b'` and `d'`. There's not much one can do about this, but at least one can say that the sign for `b'` will match the sign for `d'`. ALGORITHM: Proposition 9 of "Efficient Computation of p-adic Heights" (David Harvey, to appear in LMS JCM). Complexity is soft-`O(\log L \log m + \log^2 m)`. AUTHORS: - David Harvey (2008-01): replaced _DivPolyContext with _multiply_point EXAMPLES: 37a has trivial Tamagawa numbers so all points have nonsingular reduction at all primes:: sage: E = EllipticCurve("37a") sage: P = E([0, -1]); P (0 : -1 : 1) sage: 19*P (-59997896/67387681 : 88075171080/553185473329 : 1) sage: R = Integers(625) sage: from sage.schemes.elliptic_curves.padics import _multiply_point sage: _multiply_point(E, R, P, 19) (229, 170, 541) sage: -59997896 % 625 229 sage: -88075171080 % 625 # note sign is flipped 170 sage: -67387681.sqrt() % 625 # sign is flipped here too 541 Trivial cases (trac 3632):: sage: _multiply_point(E, R, P, 1) (0, 624, 1) sage: _multiply_point(E, R, 19*P, 1) (229, 455, 84) sage: (170 + 455) % 625 # note the sign did not flip here 0 sage: (541 + 84) % 625 0 Test over a range of `n` for a single curve with fairly random coefficients:: sage: R = Integers(625) sage: E = EllipticCurve([4, -11, 17, -8, -10]) sage: P = E.gens()[0] * LCM(E.tamagawa_numbers()) sage: from sage.schemes.elliptic_curves.padics import _multiply_point sage: Q = P sage: for n in range(1, 25): ... naive = R(Q[0].numerator()), \ ... R(Q[1].numerator()), \ ... R(Q[0].denominator().sqrt()) ... triple = _multiply_point(E, R, P, n) ... assert (triple[0] == naive[0]) and ( \ ... (triple[1] == naive[1] and triple[2] == naive[2]) or \ ... (triple[1] == -naive[1] and triple[2] == -naive[2])), \ ... "_multiply_point() gave an incorrect answer" ... Q = Q + P """ assert m >= 1 alpha = R(P[0].numerator()) beta = R(P[1].numerator()) d = R(P[0].denominator().sqrt()) if m == 1: return alpha, beta, d a1 = R(E.a1()) * d a3 = R(E.a3()) * d**3 b2 = R(E.b2()) * d**2 b4 = R(E.b4()) * d**4 b6 = R(E.b6()) * d**6 b8 = R(E.b8()) * d**8 B4 = 6*alpha**2 + b2*alpha + b4 B6 = 4*alpha**3 + b2*alpha**2 + 2*b4*alpha + b6 B6_sqr = B6*B6 B8 = 3*alpha**4 + b2*alpha**3 + 3*b4*alpha**2 + 3*b6*alpha + b8 T = 2*beta + a1*alpha + a3 # make a list of disjoint intervals [a[i], b[i]) such that we need to # compute g(k) for all a[i] <= k <= b[i] for each i intervals = [] interval = (m - 2, m + 3) while interval[0] < interval[1]: intervals.append(interval) interval = max((interval[0] - 3) >> 1, 0), \ min((interval[1] + 5) >> 1, interval[0]) # now walk through list and compute g(k) g = {0 : R(0), 1 : R(1), 2 : R(-1), 3 : B8, 4 : B6**2 - B4*B8} last = [0, 1, 2, 3, 4] # last few k for i in reversed(intervals): k = i[0] while k < i[1]: if k > 4: j = k >> 1 if k & 1: t1 = g[j] t2 = g[j+1] prod1 = g[j+2] * t1*t1*t1 prod2 = g[j-1] * t2*t2*t2 g[k] = prod1 - B6_sqr * prod2 if j & 1 else B6_sqr * prod1 - prod2 else: t1 = g[j-1] t2 = g[j+1] g[k] = g[j] * (g[j-2] * t2*t2 - g[j+2] * t1*t1) k = k + 1 if m & 1: psi_m = g[m] psi_m_m1 = g[m-1] * T psi_m_p1 = g[m+1] * T else: psi_m = g[m] * T psi_m_m1 = g[m-1] psi_m_p1 = g[m+1] theta = alpha * psi_m * psi_m - psi_m_m1 * psi_m_p1 t1 = g[m-2] * g[m+1] * g[m+1] - g[m+2] * g[m-1] * g[m-1] if m & 1: t1 = t1 * T omega = (t1 + (a1 * theta + a3 * psi_m * psi_m) * psi_m) / -2 return theta, omega, psi_m * d def padic_height(self, p, prec=20, sigma=None, check_hypotheses=True): r""" Computes the cyclotomic p-adic height. The equation of the curve must be minimal at `p`. INPUT: - ``p`` - prime = 5 for which the curve has semi-stable reduction - ``prec`` - integer = 1, desired precision of result - ``sigma`` - precomputed value of sigma. If not supplied, this function will call padic_sigma to compute it. - ``check_hypotheses`` - boolean, whether to check that this is a curve for which the p-adic height makes sense OUTPUT: A function that accepts two parameters: - a Q-rational point on the curve whose height should be computed - optional boolean flag 'check': if False, it skips some input checking, and returns the p-adic height of that point to the desired precision. - The normalization (sign and a factor 1/2 with respect to some other normalizations that appear in the literature) is chosen in such a way as to make the p-adic Birch Swinnerton-Dyer conjecture hold as stated in [Mazur-Tate-Teitelbaum]. AUTHORS: - Jennifer Balakrishnan: original code developed at the 2006 MSRI graduate workshop on modular forms - David Harvey (2006-09-13): integrated into Sage, optimised to speed up repeated evaluations of the returned height function, addressed some thorny precision questions - David Harvey (2006-09-30): rewrote to use division polynomials for computing denominator of `nP`. - David Harvey (2007-02): cleaned up according to algorithms in "Efficient Computation of p-adic Heights" - Chris Wuthrich (2007-05): added supersingular and multiplicative heights EXAMPLES:: sage: E = EllipticCurve("37a") sage: P = E.gens()[0] sage: h = E.padic_height(5, 10) sage: h(P) 5 + 5^2 + 5^3 + 3*5^6 + 4*5^7 + 5^9 + O(5^10) An anomalous case:: sage: h = E.padic_height(53, 10) sage: h(P) 26*53^-1 + 30 + 20*53 + 47*53^2 + 10*53^3 + 32*53^4 + 9*53^5 + 22*53^6 + 35*53^7 + 30*53^8 + 17*53^9 + O(53^10) Boundary case:: sage: E.padic_height(5, 3)(P) 5 + 5^2 + O(5^3) A case that works the division polynomial code a little harder:: sage: E.padic_height(5, 10)(5*P) 5^3 + 5^4 + 5^5 + 3*5^8 + 4*5^9 + O(5^10) Check that answers agree over a range of precisions:: sage: max_prec = 30 # make sure we get past p^2 # long time sage: full = E.padic_height(5, max_prec)(P) # long time sage: for prec in range(1, max_prec): # long time ... assert E.padic_height(5, prec)(P) == full # long time A supersingular prime for a curve:: sage: E = EllipticCurve('37a') sage: E.is_supersingular(3) True sage: h = E.padic_height(3, 5) sage: h(E.gens()[0]) (3 + 3^3 + O(3^6), 2*3^2 + 3^3 + 3^4 + 3^5 + 2*3^6 + O(3^7)) sage: E.padic_regulator(5) 5 + 5^2 + 5^3 + 3*5^6 + 4*5^7 + 5^9 + 5^10 + 3*5^11 + 3*5^12 + 5^13 + 4*5^14 + 5^15 + 2*5^16 + 5^17 + 2*5^18 + 4*5^19 + O(5^20) sage: E.padic_regulator(3, 5) (3 + 2*3^2 + 3^3 + O(3^4), 3^2 + 2*3^3 + 3^4 + O(3^5)) A torsion point in both the good and supersingular cases:: sage: E = EllipticCurve('11a') sage: P = E.torsion_subgroup().gen(0).element(); P (5 : 5 : 1) sage: h = E.padic_height(19, 5) sage: h(P) 0 sage: h = E.padic_height(5, 5) sage: h(P) 0 The result is not dependent on the model for the curve:: sage: E = EllipticCurve([0,0,0,0,2^12*17]) sage: Em = E.minimal_model() sage: P = E.gens()[0] sage: Pm = Em.gens()[0] sage: h = E.padic_height(7) sage: hm = Em.padic_height(7) sage: h(P) == hm(Pm) True """ if check_hypotheses: if not p.is_prime(): raise ValueError, "p = (%s) must be prime"%p if p == 2: raise ValueError, "p must be odd" # todo if self.conductor() % (p**2) == 0: raise ArithmeticError, "p must be a semi-stable prime" prec = int(prec) if prec < 1: raise ValueError, "prec (=%s) must be at least 1" % prec if self.conductor() % p == 0: Eq = self.tate_curve(p,prec=prec) return Eq.height(prec=prec) elif self.ap(p) % p == 0: lp = self.padic_lseries(p) return lp.Dp_valued_height(prec=prec) # else good ordinary case # For notation and definitions, see "Efficient Computation of # p-adic Heights", David Harvey (unpublished) n1 = self.change_ring(rings.GF(p)).cardinality() n2 = arith.LCM(self.tamagawa_numbers()) n = arith.LCM(n1, n2) m = int(n / n2) adjusted_prec = prec + 2 * arith.valuation(n, p) # this is M' R = rings.Integers(p ** adjusted_prec) if sigma is None: sigma = self.padic_sigma(p, adjusted_prec, check_hypotheses=False) # K is the field for the final result K = Qp(p, prec=adjusted_prec-1) E = self def height(P, check=True): if P.is_finite_order(): return K(0) if check: assert P.curve() == E, "the point P must lie on the curve " \ "from which the height function was created" Q = n2 * P alpha, beta, d = _multiply_point(E, R, Q, m) assert beta.lift() % p != 0, "beta should be a unit!" assert d.lift() % p == 0, "d should not be a unit!" t = -d * alpha / beta total = R(1) t_power = t for k in range(2, adjusted_prec + 1): total = total + t_power * sigma[k].lift() t_power = t_power * t total = (-alpha / beta) * total L = Qp(p, prec=adjusted_prec) total = L(total.lift(), adjusted_prec) # yuck... get rid of this lift! # changed sign to make it correct for p-adic bsd answer = -total.log() * 2 / n**2 if check: assert answer.precision_absolute() >= prec, "we should have got an " \ "answer with precision at least prec, but we didn't." return K(answer) # (man... I love python's local function definitions...) return height def padic_height_via_multiply(self, p, prec=20, E2=None, check_hypotheses=True): r""" Computes the cyclotomic p-adic height. The equation of the curve must be minimal at `p`. INPUT: - ``p`` - prime = 5 for which the curve has good ordinary reduction - ``prec`` - integer = 2, desired precision of result - ``E2`` - precomputed value of E2. If not supplied, this function will call padic_E2 to compute it. The value supplied must be correct mod `p^(prec-2)` (or slightly higher in the anomalous case; see the code for details). - ``check_hypotheses`` - boolean, whether to check that this is a curve for which the p-adic height makes sense OUTPUT: A function that accepts two parameters: - a Q-rational point on the curve whose height should be computed - optional boolean flag 'check': if False, it skips some input checking, and returns the p-adic height of that point to the desired precision. - The normalization (sign and a factor 1/2 with respect to some other normalizations that appear in the literature) is chosen in such a way as to make the p-adic Birch Swinnerton-Dyer conjecture hold as stated in [Mazur-Tate-Teitelbaum]. AUTHORS: - David Harvey (2008-01): based on the padic_height() function, using the algorithm of"Computing p-adic heights via point multiplication" EXAMPLES:: sage: E = EllipticCurve("37a") sage: P = E.gens()[0] sage: h = E.padic_height_via_multiply(5, 10) sage: h(P) 5 + 5^2 + 5^3 + 3*5^6 + 4*5^7 + 5^9 + O(5^10) An anomalous case:: sage: h = E.padic_height_via_multiply(53, 10) sage: h(P) 26*53^-1 + 30 + 20*53 + 47*53^2 + 10*53^3 + 32*53^4 + 9*53^5 + 22*53^6 + 35*53^7 + 30*53^8 + 17*53^9 + O(53^10) Supply the value of E2 manually:: sage: E2 = E.padic_E2(5, 8) sage: E2 2 + 4*5 + 2*5^3 + 5^4 + 3*5^5 + 2*5^6 + O(5^8) sage: h = E.padic_height_via_multiply(5, 10, E2=E2) sage: h(P) 5 + 5^2 + 5^3 + 3*5^6 + 4*5^7 + 5^9 + O(5^10) Boundary case:: sage: E.padic_height_via_multiply(5, 3)(P) 5 + 5^2 + O(5^3) Check that answers agree over a range of precisions:: sage: max_prec = 30 # make sure we get past p^2 # long time sage: full = E.padic_height(5, max_prec)(P) # long time sage: for prec in range(2, max_prec): # long time ... assert E.padic_height_via_multiply(5, prec)(P) == full # long time """ if check_hypotheses: if not p.is_prime(): raise ValueError, "p = (%s) must be prime"%p if p == 2: raise ValueError, "p must be odd" # todo if self.conductor() % p == 0: raise ArithmeticError, "must have good reduction at p" if self.ap(p) % p == 0: raise ArithmeticError, "must be ordinary at p" prec = int(prec) if prec < 1: raise ValueError, "prec (=%s) must be at least 1" % prec # For notation and definitions, see ``Computing p-adic heights via point # multiplication'' (David Harvey, still in draft form) n1 = self.change_ring(rings.GF(p)).cardinality() n2 = arith.LCM(self.tamagawa_numbers()) n = arith.LCM(n1, n2) m = int(n / n2) lamb = int(math.floor(math.sqrt(prec))) adjusted_prec = prec + 2 * arith.valuation(n, p) # this is M' R = rings.Integers(p ** (adjusted_prec + 2*lamb)) sigma = self.padic_sigma_truncated(p, N=adjusted_prec, E2=E2, lamb=lamb) # K is the field for the final result K = Qp(p, prec=adjusted_prec-1) E = self def height(P, check=True): if P.is_finite_order(): return K(0) if check: assert P.curve() == E, "the point P must lie on the curve " \ "from which the height function was created" Q = n2 * P alpha, beta, d = _multiply_point(E, R, Q, m * p**lamb) assert beta.lift() % p != 0, "beta should be a unit!" assert d.lift() % p == 0, "d should not be a unit!" t = -d * alpha / beta total = R(1) t_power = t for k in range(2, sigma.prec()): total = total + t_power * sigma[k].lift() t_power = t_power * t total = (-alpha / beta) * total L = Qp(p, prec=adjusted_prec + 2*lamb) total = L(total.lift(), adjusted_prec + 2*lamb) # changed sign to make it correct for p-adic bsd answer = -total.log() * 2 / (n * p**lamb)**2 if check: assert answer.precision_absolute() >= prec, "we should have got an " \ "answer with precision at least prec, but we didn't." return K(answer) # (man... I love python's local function definitions...) return height def padic_sigma(self, p, N=20, E2=None, check=False, check_hypotheses=True): r""" Computes the p-adic sigma function with respect to the standard invariant differential `dx/(2y + a_1 x + a_3)`, as defined by Mazur and Tate, as a power series in the usual uniformiser `t` at the origin. The equation of the curve must be minimal at `p`. INPUT: - ``p`` - prime = 5 for which the curve has good ordinary reduction - ``N`` - integer = 1, indicates precision of result; see OUTPUT section for description - ``E2`` - precomputed value of E2. If not supplied, this function will call padic_E2 to compute it. The value supplied must be correct mod `p^{N-2}`. - ``check`` - boolean, whether to perform a consistency check (i.e. verify that the computed sigma satisfies the defining - ``differential equation`` - note that this does NOT guarantee correctness of all the returned digits, but it comes pretty close :-)) - ``check_hypotheses`` - boolean, whether to check that this is a curve for which the p-adic sigma function makes sense OUTPUT: A power series `t + \cdots` with coefficients in `\ZZ_p`. The output series will be truncated at `O(t^{N+1})`, and the coefficient of `t^n` for `n \geq 1` will be correct to precision `O(p^{N-n+1})`. In practice this means the following. If `t_0 = p^k u`, where `u` is a `p`-adic unit with at least `N` digits of precision, and `k \geq 1`, then the returned series may be used to compute `\sigma(t_0)` correctly modulo `p^{N+k}` (i.e. with `N` correct `p`-adic digits). ALGORITHM: Described in "Efficient Computation of p-adic Heights" (David Harvey), which is basically an optimised version of the algorithm from "p-adic Heights and Log Convergence" (Mazur, Stein, Tate). Running time is soft-`O(N^2 \log p)`, plus whatever time is necessary to compute `E_2`. AUTHORS: - David Harvey (2006-09-12) - David Harvey (2007-02): rewrote EXAMPLES:: sage: EllipticCurve([-1, 1/4]).padic_sigma(5, 10) O(5^11) + (1 + O(5^10))*t + O(5^9)*t^2 + (3 + 2*5^2 + 3*5^3 + 3*5^6 + 4*5^7 + O(5^8))*t^3 + O(5^7)*t^4 + (2 + 4*5^2 + 4*5^3 + 5^4 + 5^5 + O(5^6))*t^5 + O(5^5)*t^6 + (2 + 2*5 + 5^2 + 4*5^3 + O(5^4))*t^7 + O(5^3)*t^8 + (1 + 2*5 + O(5^2))*t^9 + O(5)*t^10 + O(t^11) Run it with a consistency check:: sage: EllipticCurve("37a").padic_sigma(5, 10, check=True) O(5^11) + (1 + O(5^10))*t + O(5^9)*t^2 + (3 + 2*5^2 + 3*5^3 + 3*5^6 + 4*5^7 + O(5^8))*t^3 + (3 + 2*5 + 2*5^2 + 2*5^3 + 2*5^4 + 2*5^5 + 2*5^6 + O(5^7))*t^4 + (2 + 4*5^2 + 4*5^3 + 5^4 + 5^5 + O(5^6))*t^5 + (2 + 3*5 + 5^4 + O(5^5))*t^6 + (4 + 3*5 + 2*5^2 + O(5^4))*t^7 + (2 + 3*5 + 2*5^2 + O(5^3))*t^8 + (4*5 + O(5^2))*t^9 + (1 + O(5))*t^10 + O(t^11) Boundary cases:: sage: EllipticCurve([1, 1, 1, 1, 1]).padic_sigma(5, 1) (1 + O(5))*t + O(t^2) sage: EllipticCurve([1, 1, 1, 1, 1]).padic_sigma(5, 2) (1 + O(5^2))*t + (3 + O(5))*t^2 + O(t^3) Supply your very own value of E2:: sage: X = EllipticCurve("37a") sage: my_E2 = X.padic_E2(5, 8) sage: my_E2 = my_E2 + 5**5 # oops!!! sage: X.padic_sigma(5, 10, E2=my_E2) O(5^11) + (1 + O(5^10))*t + O(5^9)*t^2 + (3 + 2*5^2 + 3*5^3 + 4*5^5 + 2*5^6 + 3*5^7 + O(5^8))*t^3 + (3 + 2*5 + 2*5^2 + 2*5^3 + 2*5^4 + 2*5^5 + 2*5^6 + O(5^7))*t^4 + (2 + 4*5^2 + 4*5^3 + 5^4 + 3*5^5 + O(5^6))*t^5 + (2 + 3*5 + 5^4 + O(5^5))*t^6 + (4 + 3*5 + 2*5^2 + O(5^4))*t^7 + (2 + 3*5 + 2*5^2 + O(5^3))*t^8 + (4*5 + O(5^2))*t^9 + (1 + O(5))*t^10 + O(t^11) Check that sigma is "weight 1". :: sage: f = EllipticCurve([-1, 3]).padic_sigma(5, 10) sage: g = EllipticCurve([-1*(2**4), 3*(2**6)]).padic_sigma(5, 10) sage: t = f.parent().gen() sage: f(2*t)/2 (1 + O(5^10))*t + (4 + 3*5 + 3*5^2 + 3*5^3 + 4*5^4 + 4*5^5 + 3*5^6 + 5^7 + O(5^8))*t^3 + (3 + 3*5^2 + 5^4 + 2*5^5 + O(5^6))*t^5 + (4 + 5 + 3*5^3 + O(5^4))*t^7 + (4 + 2*5 + O(5^2))*t^9 + O(5)*t^10 + O(t^11) sage: g O(5^11) + (1 + O(5^10))*t + O(5^9)*t^2 + (4 + 3*5 + 3*5^2 + 3*5^3 + 4*5^4 + 4*5^5 + 3*5^6 + 5^7 + O(5^8))*t^3 + O(5^7)*t^4 + (3 + 3*5^2 + 5^4 + 2*5^5 + O(5^6))*t^5 + O(5^5)*t^6 + (4 + 5 + 3*5^3 + O(5^4))*t^7 + O(5^3)*t^8 + (4 + 2*5 + O(5^2))*t^9 + O(5)*t^10 + O(t^11) sage: f(2*t)/2 -g O(t^11) Test that it returns consistent results over a range of precision:: sage: max_N = 30 # get up to at least p^2 # long time sage: E = EllipticCurve([1, 1, 1, 1, 1]) # long time sage: p = 5 # long time sage: E2 = E.padic_E2(5, max_N) # long time sage: max_sigma = E.padic_sigma(p, max_N, E2=E2) # long time sage: for N in range(3, max_N): # long time ... sigma = E.padic_sigma(p, N, E2=E2) # long time ... assert sigma == max_sigma """ if check_hypotheses: p = __check_padic_hypotheses(self, p) # todo: implement the p == 3 case # NOTE: If we ever implement p == 3, it's necessary to check over # the precision loss estimates (below) very carefully; I think it # may become necessary to compute E2 to an even higher precision. if p < 5: raise NotImplementedError, "p (=%s) must be at least 5" % p N = int(N) if N <= 2: # a few special cases for small N if N < 1: raise ValueError, "N (=%s) must be at least 1" % prec if N == 1: # return simply t + O(t^2) K = Qp(p, 2) return PowerSeriesRing(K, "t")([K(0), K(1, 1)], prec=2) if N == 2: # return t + a_1/2 t^2 + O(t^3) K = Qp(p, 3) return PowerSeriesRing(K, "t")([K(0), K(1, 2), K(self.a1()/2, 1)], prec=3) if self.discriminant().valuation(p) != 0: raise NotImplementedError, "equation of curve must be minimal at p" if E2 is None: E2 = self.padic_E2(p, N-2, check_hypotheses=False) elif E2.precision_absolute() < N-2: raise ValueError, "supplied E2 has insufficient precision" QQt = LaurentSeriesRing(RationalField(), "x") R = rings.Integers(p**(N-2)) X = self.change_ring(R) c = (X.a1()**2 + 4*X.a2() - R(E2)) / 12 f = X.formal_group().differential(N+2) # f = 1 + ... + O(t^{N+2}) x = X.formal_group().x(N) # x = t^{-2} + ... + O(t^N) Rt = x.parent() A = (x + c) * f # do integral over QQ, to avoid divisions by p A = Rt(QQt(A).integral()) A = (-X.a1()/2 - A) * f # Convert to a power series and remove the -1/x term. # Also we artificially bump up the accuracy from N-2 to to N-1 digits; # the constant term needs to be known to N-1 digits, so we compute # it directly assert A.valuation() == -1 and A[-1] == 1 A = A - A.parent().gen() ** (-1) A = A.power_series().list() R = rings.Integers(p**(N-1)) A = [R(u) for u in A] A[0] = self.change_ring(R).a1()/2 # fix constant term A = PowerSeriesRing(R, "x")(A, len(A)) theta = _brent(A, p, N) sigma = theta * theta.parent().gen() # Convert the answer to power series over p-adics; drop the precision # of the $t^k$ coefficient to $p^(N-k+1)$. # [Note: there are actually more digits available, but it's a bit # tricky to figure out exactly how many, and we only need $p^(N-k+1)$ # for p-adic height purposes anyway] K = rings.pAdicField(p, N + 1) sigma = sigma.padded_list(N+1) sigma[0] = K(0, N +1) sigma[1] = K(1, N) for n in range(2, N+1): sigma[n] = K(sigma[n].lift(), N - n + 1) S = rings.PowerSeriesRing(K, "t", N+1) sigma = S(sigma, N+1) # if requested, check that sigma satisfies the appropriate # differential equation if check: R = rings.Integers(p**N) X = self.change_ring(R) x = X.formal_group().x(N+5) # few extra terms for safety f = X.formal_group().differential(N+5) c = (X.a1()**2 + 4*X.a2() - R(E2)) / 12 # convert sigma to be over Z/p^N s = f.parent()(sigma) sinv = s**(-1) finv = f**(-1) # apply differential equation temp = (s.derivative() * sinv * finv).derivative() * finv + c + x # coefficient of t^k in the result should be zero mod p^(N-k-2) for k in range(N-2): assert temp[k].lift().valuation(p) >= N - k - 2, \ "sigma correctness check failed!" return sigma def padic_sigma_truncated(self, p, N=20, lamb=0, E2=None, check_hypotheses=True): r""" Computes the p-adic sigma function with respect to the standard invariant differential `dx/(2y + a_1 x + a_3)`, as defined by Mazur and Tate, as a power series in the usual uniformiser `t` at the origin. The equation of the curve must be minimal at `p`. This function differs from padic_sigma() in the precision profile of the returned power series; see OUTPUT below. INPUT: - ``p`` - prime = 5 for which the curve has good ordinary reduction - ``N`` - integer = 2, indicates precision of result; see OUTPUT section for description - ``lamb`` - integer = 0, see OUTPUT section for description - ``E2`` - precomputed value of E2. If not supplied, this function will call padic_E2 to compute it. The value supplied must be correct mod `p^{N-2}`. - ``check_hypotheses`` - boolean, whether to check that this is a curve for which the p-adic sigma function makes sense OUTPUT: A power series `t + \cdots` with coefficients in `\ZZ_p`. The coefficient of `t^j` for `j \geq 1` will be correct to precision `O(p^{N - 2 + (3 - j)(lamb + 1)})`. ALGORITHM: Described in "Efficient Computation of p-adic Heights" (David Harvey, to appear in LMS JCM), which is basically an optimised version of the algorithm from "p-adic Heights and Log Convergence" (Mazur, Stein, Tate), and "Computing p-adic heights via point multiplication" (David Harvey, still draft form). Running time is soft-`O(N^2 \lambda^{-1} \log p)`, plus whatever time is necessary to compute `E_2`. AUTHOR: - David Harvey (2008-01): wrote based on previous padic_sigma function EXAMPLES:: sage: E = EllipticCurve([-1, 1/4]) sage: E.padic_sigma_truncated(5, 10) O(5^11) + (1 + O(5^10))*t + O(5^9)*t^2 + (3 + 2*5^2 + 3*5^3 + 3*5^6 + 4*5^7 + O(5^8))*t^3 + O(5^7)*t^4 + (2 + 4*5^2 + 4*5^3 + 5^4 + 5^5 + O(5^6))*t^5 + O(5^5)*t^6 + (2 + 2*5 + 5^2 + 4*5^3 + O(5^4))*t^7 + O(5^3)*t^8 + (1 + 2*5 + O(5^2))*t^9 + O(5)*t^10 + O(t^11) Note the precision of the `t^3` coefficient depends only on `N`, not on lamb:: sage: E.padic_sigma_truncated(5, 10, lamb=2) O(5^17) + (1 + O(5^14))*t + O(5^11)*t^2 + (3 + 2*5^2 + 3*5^3 + 3*5^6 + 4*5^7 + O(5^8))*t^3 + O(5^5)*t^4 + (2 + O(5^2))*t^5 + O(t^6) Compare against plain padic_sigma() function over a dense range of N and lamb :: sage: E = EllipticCurve([1, 2, 3, 4, 7]) # long time sage: E2 = E.padic_E2(5, 50) # long time sage: for N in range(2, 10): # long time ... for lamb in range(10): # long time ... correct = E.padic_sigma(5, N + 3*lamb, E2=E2) # long time ... compare = E.padic_sigma_truncated(5, N=N, lamb=lamb, E2=E2) # long time ... assert compare == correct # long time """ if check_hypotheses: p = __check_padic_hypotheses(self, p) # todo: implement the p == 3 case # NOTE: If we ever implement p == 3, it's necessary to check over # the precision loss estimates (below) very carefully; I think it # may become necessary to compute E2 to an even higher precision. if p < 5: raise NotImplementedError, "p (=%s) must be at least 5" % p N = int(N) lamb = int(lamb) if lamb < 0: raise ValueError, "lamb (=%s) must be at least 0" % lamb # a few special cases for small N if N <= 1: raise ValueError, "N (=%s) must be at least 2" % N if N == 2: # return t + a_1/2 t^2 + O(t^3) K = Qp(p, 3*(lamb+1)) return PowerSeriesRing(K, "t")([K(0), K(1, 2*(lamb+1)), K(self.a1()/2, lamb+1)], prec=3) if self.discriminant().valuation(p) != 0: raise NotImplementedError, "equation of curve must be minimal at p" if E2 is None: E2 = self.padic_E2(p, N-2, check_hypotheses=False) elif E2.precision_absolute() < N-2: raise ValueError, "supplied E2 has insufficient precision" # The main part of the algorithm is exactly the same as # for padic_sigma(), but we truncate all the series earlier. # Want the answer O(t^(trunc+1)) instead of O(t^(N+1)) like in padic_sigma(). trunc = (Integer(N-2) / (lamb + 1)).ceil() + 2 QQt = LaurentSeriesRing(RationalField(), "x") R = rings.Integers(p**(N-2)) X = self.change_ring(R) c = (X.a1()**2 + 4*X.a2() - R(E2)) / 12 f = X.formal_group().differential(trunc+2) # f = 1 + ... + O(t^{trunc+2}) x = X.formal_group().x(trunc) # x = t^{-2} + ... + O(t^trunc) Rt = x.parent() A = (x + c) * f # do integral over QQ, to avoid divisions by p A = Rt(QQt(A).integral()) A = (-X.a1()/2 - A) * f # Convert to a power series and remove the -1/x term. # Also we artificially bump up the accuracy from N-2 to N-1+lamb digits; # the constant term needs to be known to N-1+lamb digits, so we compute # it directly assert A.valuation() == -1 and A[-1] == 1 A = A - A.parent().gen() ** (-1) A = A.power_series().list() R = rings.Integers(p**(N-1+lamb)) A = [R(u) for u in A] A[0] = self.change_ring(R).a1()/2 # fix constant term A = PowerSeriesRing(R, "x")(A, len(A)) theta = _brent(A, p, trunc) sigma = theta * theta.parent().gen() # Convert the answer to power series over p-adics; drop the precision # of the $t^j$ coefficient to $p^{N - 2 + (3 - j)(lamb + 1)})$. K = rings.pAdicField(p, N - 2 + 3*(lamb+1)) sigma = sigma.padded_list(trunc+1) sigma[0] = K(0, N - 2 + 3*(lamb+1)) sigma[1] = K(1, N - 2 + 2*(lamb+1)) for j in range(2, trunc+1): sigma[j] = K(sigma[j].lift(), N - 2 + (3 - j)*(lamb+1)) S = rings.PowerSeriesRing(K, "t", trunc + 1) sigma = S(sigma, trunc+1) return sigma def padic_E2(self, p, prec=20, check=False, check_hypotheses=True, algorithm="auto"): r""" Returns the value of the `p`-adic modular form `E2` for `(E, \omega)` where `\omega` is the usual invariant differential `dx/(2y + a_1 x + a_3)`. INPUT: - ``p`` - prime (= 5) for which `E` is good and ordinary - ``prec`` - (relative) p-adic precision (= 1) for result - ``check`` - boolean, whether to perform a consistency check. This will slow down the computation by a constant factor 2. (The consistency check is to compute the whole matrix of frobenius on Monsky-Washnitzer cohomology, and verify that its trace is correct to the specified precision. Otherwise, the trace is used to compute one column from the other one (possibly after a change of basis).) - ``check_hypotheses`` - boolean, whether to check that this is a curve for which the p-adic sigma function makes sense - ``algorithm`` - one of "standard", "sqrtp", or "auto". This selects which version of Kedlaya's algorithm is used. The "standard" one is the one described in Kedlaya's paper. The "sqrtp" one has better performance for large `p`, but only works when `p > 6N` (`N=` prec). The "auto" option selects "sqrtp" whenever possible. Note that if the "sqrtp" algorithm is used, a consistency check will automatically be applied, regardless of the setting of the "check" flag. OUTPUT: p-adic number to precision prec .. note:: If the discriminant of the curve has nonzero valuation at p, then the result will not be returned mod `p^\text{prec}`, but it still *will* have prec *digits* of precision. TODO: - Once we have a better implementation of the "standard" algorithm, the algorithm selection strategy for "auto" needs to be revisited. AUTHORS: - David Harvey (2006-09-01): partly based on code written by Robert Bradshaw at the MSRI 2006 modular forms workshop ACKNOWLEDGMENT: - discussion with Eyal Goren that led to the trace trick. EXAMPLES: Here is the example discussed in the paper "Computation of p-adic Heights and Log Convergence" (Mazur, Stein, Tate):: sage: EllipticCurve([-1, 1/4]).padic_E2(5) 2 + 4*5 + 2*5^3 + 5^4 + 3*5^5 + 2*5^6 + 5^8 + 3*5^9 + 4*5^10 + 2*5^11 + 2*5^12 + 2*5^14 + 3*5^15 + 3*5^16 + 3*5^17 + 4*5^18 + 2*5^19 + O(5^20) Let's try to higher precision (this is the same answer the MAGMA implementation gives):: sage: EllipticCurve([-1, 1/4]).padic_E2(5, 100) 2 + 4*5 + 2*5^3 + 5^4 + 3*5^5 + 2*5^6 + 5^8 + 3*5^9 + 4*5^10 + 2*5^11 + 2*5^12 + 2*5^14 + 3*5^15 + 3*5^16 + 3*5^17 + 4*5^18 + 2*5^19 + 4*5^20 + 5^21 + 4*5^22 + 2*5^23 + 3*5^24 + 3*5^26 + 2*5^27 + 3*5^28 + 2*5^30 + 5^31 + 4*5^33 + 3*5^34 + 4*5^35 + 5^36 + 4*5^37 + 4*5^38 + 3*5^39 + 4*5^41 + 2*5^42 + 3*5^43 + 2*5^44 + 2*5^48 + 3*5^49 + 4*5^50 + 2*5^51 + 5^52 + 4*5^53 + 4*5^54 + 3*5^55 + 2*5^56 + 3*5^57 + 4*5^58 + 4*5^59 + 5^60 + 3*5^61 + 5^62 + 4*5^63 + 5^65 + 3*5^66 + 2*5^67 + 5^69 + 2*5^70 + 3*5^71 + 3*5^72 + 5^74 + 5^75 + 5^76 + 3*5^77 + 4*5^78 + 4*5^79 + 2*5^80 + 3*5^81 + 5^82 + 5^83 + 4*5^84 + 3*5^85 + 2*5^86 + 3*5^87 + 5^88 + 2*5^89 + 4*5^90 + 4*5^92 + 3*5^93 + 4*5^94 + 3*5^95 + 2*5^96 + 4*5^97 + 4*5^98 + 2*5^99 + O(5^100) Check it works at low precision too:: sage: EllipticCurve([-1, 1/4]).padic_E2(5, 1) 2 + O(5) sage: EllipticCurve([-1, 1/4]).padic_E2(5, 2) 2 + 4*5 + O(5^2) sage: EllipticCurve([-1, 1/4]).padic_E2(5, 3) 2 + 4*5 + O(5^3) TODO: With the old(-er), i.e., = sage-2.4 p-adics we got `5 + O(5^2)` as output, i.e., relative precision 1, but with the newer p-adics we get relative precision 0 and absolute precision 1. :: sage: EllipticCurve([1, 1, 1, 1, 1]).padic_E2(5, 1) O(5) Check it works for different models of the same curve (37a), even when the discriminant changes by a power of p (note that E2 depends on the differential too, which is why it gets scaled in some of the examples below):: sage: X1 = EllipticCurve([-1, 1/4]) sage: X1.j_invariant(), X1.discriminant() (110592/37, 37) sage: X1.padic_E2(5, 10) 2 + 4*5 + 2*5^3 + 5^4 + 3*5^5 + 2*5^6 + 5^8 + 3*5^9 + O(5^10) :: sage: X2 = EllipticCurve([0, 0, 1, -1, 0]) sage: X2.j_invariant(), X2.discriminant() (110592/37, 37) sage: X2.padic_E2(5, 10) 2 + 4*5 + 2*5^3 + 5^4 + 3*5^5 + 2*5^6 + 5^8 + 3*5^9 + O(5^10) :: sage: X3 = EllipticCurve([-1*(2**4), 1/4*(2**6)]) sage: X3.j_invariant(), X3.discriminant() / 2**12 (110592/37, 37) sage: 2**(-2) * X3.padic_E2(5, 10) 2 + 4*5 + 2*5^3 + 5^4 + 3*5^5 + 2*5^6 + 5^8 + 3*5^9 + O(5^10) :: sage: X4 = EllipticCurve([-1*(7**4), 1/4*(7**6)]) sage: X4.j_invariant(), X4.discriminant() / 7**12 (110592/37, 37) sage: 7**(-2) * X4.padic_E2(5, 10) 2 + 4*5 + 2*5^3 + 5^4 + 3*5^5 + 2*5^6 + 5^8 + 3*5^9 + O(5^10) :: sage: X5 = EllipticCurve([-1*(5**4), 1/4*(5**6)]) sage: X5.j_invariant(), X5.discriminant() / 5**12 (110592/37, 37) sage: 5**(-2) * X5.padic_E2(5, 10) 2 + 4*5 + 2*5^3 + 5^4 + 3*5^5 + 2*5^6 + 5^8 + 3*5^9 + O(5^10) :: sage: X6 = EllipticCurve([-1/(5**4), 1/4/(5**6)]) sage: X6.j_invariant(), X6.discriminant() * 5**12 (110592/37, 37) sage: 5**2 * X6.padic_E2(5, 10) 2 + 4*5 + 2*5^3 + 5^4 + 3*5^5 + 2*5^6 + 5^8 + 3*5^9 + O(5^10) Test check=True vs check=False:: sage: EllipticCurve([-1, 1/4]).padic_E2(5, 1, check=False) 2 + O(5) sage: EllipticCurve([-1, 1/4]).padic_E2(5, 1, check=True) 2 + O(5) sage: EllipticCurve([-1, 1/4]).padic_E2(5, 30, check=False) 2 + 4*5 + 2*5^3 + 5^4 + 3*5^5 + 2*5^6 + 5^8 + 3*5^9 + 4*5^10 + 2*5^11 + 2*5^12 + 2*5^14 + 3*5^15 + 3*5^16 + 3*5^17 + 4*5^18 + 2*5^19 + 4*5^20 + 5^21 + 4*5^22 + 2*5^23 + 3*5^24 + 3*5^26 + 2*5^27 + 3*5^28 + O(5^30) sage: EllipticCurve([-1, 1/4]).padic_E2(5, 30, check=True) 2 + 4*5 + 2*5^3 + 5^4 + 3*5^5 + 2*5^6 + 5^8 + 3*5^9 + 4*5^10 + 2*5^11 + 2*5^12 + 2*5^14 + 3*5^15 + 3*5^16 + 3*5^17 + 4*5^18 + 2*5^19 + 4*5^20 + 5^21 + 4*5^22 + 2*5^23 + 3*5^24 + 3*5^26 + 2*5^27 + 3*5^28 + O(5^30) Here's one using the `p^{1/2}` algorithm:: sage: EllipticCurve([-1, 1/4]).padic_E2(3001, 3, algorithm="sqrtp") 1907 + 2819*3001 + 1124*3001^2 + O(3001^3) """ if self.conductor() % p == 0: if not self.conductor() % (p**2) == 0: eq = self.tate_curve(p,prec=prec) return eq.E2(prec=prec) frob_p = self.matrix_of_frobenius(p, prec, check, check_hypotheses, algorithm).change_ring(Integers(p**prec)) frob_p_n = frob_p**prec # todo: think about the sign of this. Is it correct? output_ring = rings.pAdicField(p, prec) E2_of_X = output_ring( (-12 * frob_p_n[0,1] / frob_p_n[1,1]).lift() ) \ + O(p**prec) # Take into account the coordinate change. X = self.minimal_model().short_weierstrass_model() fudge_factor = (X.discriminant() / self.discriminant()).nth_root(6) # todo: here I should be able to write: # return E2_of_X / fudge_factor # However, there is a bug in Sage (#51 on trac) which makes this # crash sometimes when prec == 1. For example, # EllipticCurve([1, 1, 1, 1, 1]).padic_E2(5, 1) # makes it crash. I haven't figured out exactly what the bug # is yet, but for now I use the following workaround: fudge_factor_inverse = Qp(p, prec=(E2_of_X.precision_absolute() + 1))(1 / fudge_factor) return output_ring(E2_of_X * fudge_factor_inverse) def matrix_of_frobenius(self, p, prec=20, check=False, check_hypotheses=True, algorithm="auto"): r""" Returns the matrix of Frobenius on the Monsky Washnitzer cohomology of the elliptic curve. INPUT: - ``p`` - prime (= 5) for which `E` is good and ordinary - ``prec`` - (relative) `p`-adic precision for result (default 20) - ``check`` - boolean (default: False), whether to perform a consistency check. This will slow down the computation by a constant factor 2. (The consistency check is to verify that its trace is correct to the specified precision. Otherwise, the trace is used to compute one column from the other one (possibly after a change of basis).) - ``check_hypotheses`` - boolean, whether to check that this is a curve for which the `p`-adic sigma function makes sense - ``algorithm`` - one of "standard", "sqrtp", or "auto". This selects which version of Kedlaya's algorithm is used. The "standard" one is the one described in Kedlaya's paper. The "sqrtp" one has better performance for large `p`, but only works when `p > 6N` (`N=` prec). The "auto" option selects "sqrtp" whenever possible. Note that if the "sqrtp" algorithm is used, a consistency check will automatically be applied, regardless of the setting of the "check" flag. OUTPUT: a matrix of `p`-adic number to precision ``prec`` See also the documentation of padic_E2. EXAMPLES:: sage: E = EllipticCurve('37a1') sage: E.matrix_of_frobenius(7) [ 2*7 + 4*7^2 + 5*7^4 + 6*7^5 + 6*7^6 + 7^8 + 4*7^9 + 3*7^10 + 2*7^11 + 5*7^12 + 4*7^14 + 7^16 + 2*7^17 + 3*7^18 + 4*7^19 + 3*7^20 + O(7^21) 2 + 3*7 + 6*7^2 + 7^3 + 3*7^4 + 5*7^5 + 3*7^7 + 7^8 + 3*7^9 + 6*7^13 + 7^14 + 7^16 + 5*7^17 + 4*7^18 + 7^19 + O(7^20)] [ 2*7 + 3*7^2 + 7^3 + 3*7^4 + 6*7^5 + 2*7^6 + 3*7^7 + 5*7^8 + 3*7^9 + 2*7^11 + 6*7^12 + 5*7^13 + 4*7^16 + 4*7^17 + 6*7^18 + 6*7^19 + 4*7^20 + O(7^21) 6 + 4*7 + 2*7^2 + 6*7^3 + 7^4 + 6*7^7 + 5*7^8 + 2*7^9 + 3*7^10 + 4*7^11 + 7^12 + 6*7^13 + 2*7^14 + 6*7^15 + 5*7^16 + 4*7^17 + 3*7^18 + 2*7^19 + O(7^20)] sage: M = E.matrix_of_frobenius(11,prec=3); M [ 9*11 + 9*11^3 + O(11^4) 10 + 11 + O(11^3)] [ 2*11 + 11^2 + O(11^4) 6 + 11 + 10*11^2 + O(11^3)] sage: M.det() 11 + O(11^4) sage: M.trace() 6 + 10*11 + 10*11^2 + O(11^3) sage: E.ap(11) -5 """ # TODO change the basis back to the original equation. # TODO, add lots of comments like the above if check_hypotheses: p = __check_padic_hypotheses(self, p) if algorithm == "auto": algorithm = "standard" if p < 6*prec else "sqrtp" elif algorithm == "sqrtp" and p < 6*prec: raise ValueError, "sqrtp algorithm is only available when p > 6*prec" if algorithm not in ["standard", "sqrtp"]: raise ValueError, "unknown algorithm '%s'" % algorithm # todo: maybe it would be good if default prec was None, and then # it selects an appropriate precision based on how large the prime # is # todo: implement the p == 3 case if p < 5: raise NotImplementedError, "p (=%s) must be at least 5" % p prec = int(prec) if prec < 1: raise ValueError, "prec (=%s) must be at least 1" % prec # To run matrix_of_frobenius(), we need to have the equation in the # form y^2 = x^3 + ax + b, whose discriminant is invertible mod p. # When we change coordinates like this, we might scale the invariant # differential, so we need to account for this. We do this by # comparing discriminants: if the discriminants differ by u^12, # then the differentials differ by u. There's a sign ambiguity here, # but it doesn't matter because E2 changes by u^2 :-) # todo: In fact, there should be available a function that returns # exactly *which* coordinate change was used. If I had that I could # avoid the discriminant circus at the end. # todo: The following strategy won't work at all for p = 2, 3. X=self.minimal_model().short_weierstrass_model() assert X.discriminant().valuation(p) == 0, "Something's gone wrong. " \ "The discriminant of the Weierstrass model should be a unit " \ " at p." if algorithm == "standard": # Need to increase precision a little to compensate for precision # losses during the computation. (See monsky_washnitzer.py # for more details.) adjusted_prec = monsky_washnitzer.adjusted_prec(p, prec) if check: trace = None else: trace = self.ap(p) base_ring = rings.Integers(p**adjusted_prec) output_ring = rings.pAdicField(p, prec) R, x = rings.PolynomialRing(base_ring, 'x').objgen() Q = x**3 + base_ring(X.a4()) * x + base_ring(X.a6()) frob_p = monsky_washnitzer.matrix_of_frobenius( Q, p, adjusted_prec, trace) else: # algorithm == "sqrtp" p_to_prec = p**prec R = rings.PolynomialRing(Integers(), "x") Q = R([X.a6() % p_to_prec, X.a4() % p_to_prec, 0, 1]) frob_p = sage.schemes.hyperelliptic_curves.hypellfrob.hypellfrob(p, prec, Q) # let's force a trace-check since this algorithm is fairly new # and we don't want to get caught with our pants down... trace = self.ap(p) check = True # return frob_p ## why was this here ? if check: trace_of_frobenius = frob_p.trace().lift() % p**prec correct_trace = self.ap(p) % p**prec assert trace_of_frobenius == correct_trace, \ "Consistency check failed! (correct = %s, actual = %s)" % \ (correct_trace, trace_of_frobenius) return frob_p.change_ring(Zp(p, prec)) def _brent(F, p, N): r""" This is an internal function; it is used by padic_sigma(). `F` is a assumed to be a power series over `R = \ZZ/p^{N-1}\ZZ`. It solves the differential equation `G'(t)/G(t) = F(t)` using Brent's algorithm, with initial condition `G(0) = 1`. It is assumed that the solution `G` has `p`-integral coefficients. More precisely, suppose that `f(t)` is a power series with genuine `p`-adic coefficients, and suppose that `g(t)` is an exact solution to `g'(t)/g(t) = f(t)`. Let `I` be the ideal `(p^N, p^{N-1} t, \ldots, p t^{N-1}, t^N)`. The input `F(t)` should be a finite-precision approximation to `f(t)`, in the sense that `\int (F - f) dt` should lie in `I`. Then the function returns a series `G(t)` such that `(G - g)(t)` lies in `I`. Complexity should be about `O(N^2 \log^2 N \log p)`, plus some log-log factors. For more information, and a proof of the precision guarantees, see Lemma 4 in "Efficient Computation of p-adic Heights" (David Harvey). AUTHORS: - David Harvey (2007-02) EXAMPLES: Carefully test the precision guarantees:: sage: brent = sage.schemes.elliptic_curves.padics._brent sage: for p in [2, 3, 5]: ... for N in [2, 3, 10, 50]: ... R = Integers(p**(N-1)) ... Rx, x = PowerSeriesRing(R, "x").objgen() ... for _ in range(5): ... g = [R.random_element() for i in range(N)] ... g[0] = R(1) ... g = Rx(g, len(g)) ... f = g.derivative() / g ... # perturb f by something whose integral is in I ... err = [R.random_element() * p**(N-i) for i in range(N+1)] ... err = Rx(err, len(err)) ... err = err.derivative() ... F = f + err ... # run brent() and compare output modulo I ... G = brent(F, p, N) ... assert G.prec() >= N, "not enough output terms" ... err = (G - g).list() ... for i in range(len(err)): ... assert err[i].lift().valuation(p) >= (N - i), \ ... "incorrect precision output" """ Rx = F.parent() # Rx = power series ring over Z/p^{N-1} Z R = Rx.base_ring() # R = Z/p^{N-1} Z Qx = PowerSeriesRing(RationalField(), "x") # initial approximation: G = Rx(1) # loop over an appropriate increasing sequence of lengths s for s in misc.newton_method_sizes(N): # zero-extend to s terms # todo: there has to be a better way in Sage to do this... G = Rx(G.list(), s) # extend current approximation to be correct to s terms H = G.derivative() / G - F # Do the integral of H over QQ[x] to avoid division by p problems H = Rx(Qx(H).integral()) G = G * (1 - H) return G