""" Provides functions for computing division polynomials and the multiplication-by-n map on an elliptic curve. """ from typing import Tuple, Dict, Set, Mapping, Optional from public import public import warnings from sympy import symbols, FF, Poly import networkx as nx from pyecsca.ec.curve import EllipticCurve from pyecsca.ec.mod import Mod, mod from pyecsca.ec.model import ShortWeierstrassModel has_pari = False try: import cypari2 has_pari = True except ImportError: cypari2 = None def values(*ns: int) -> Mapping[int, Tuple[int, ...]]: done: Set[int] = set() vals = {} todo: Set[int] = set() todo.update(ns) while todo: val = todo.pop() if val in done: continue new: Tuple[int, ...] = () if val == -2: new = (-1,) elif val == -1: pass elif val < 0: raise ValueError(f"bad {val}") elif val in (0, 1, 2, 3): pass elif val == 4: new = (-2, 3) elif val % 2 == 0: m = (val - 2) // 2 new = (m + 1, m + 3, m, m - 1, m + 2) else: m = (val - 1) // 2 if m % 2 == 0: new = (-2, m + 2, m, m - 1, m + 1) else: new = (m + 2, m, -2, m - 1, m + 1) if new: todo.update(new) vals[val] = new done.add(val) return vals def dep_graph(*ns: int): g = nx.DiGraph() vals = values(*ns) for k, v in vals.items(): if v: for e in v: g.add_edge(k, e) else: g.add_node(k) return g, vals def dep_map(*ns: int): g, vals = dep_graph(*ns) current: Set[int] = set() ls = [] for vert in nx.lexicographical_topological_sort(g, key=lambda v: -sum(g[v].keys())): if vert in current: current.remove(vert) ls.append((vert, set(current))) current.update(vals[vert]) ls.reverse() return ls, vals @public def a_invariants(curve: EllipticCurve) -> Tuple[Mod, ...]: """ Compute the a-invariants of the curve. :param curve: The elliptic curve (only ShortWeierstrass model). :return: A tuple of 5 a-invariants (a1, a2, a3, a4, a6). """ if isinstance(curve.model, ShortWeierstrassModel): a1 = mod(0, curve.prime) a2 = mod(0, curve.prime) a3 = mod(0, curve.prime) a4 = curve.parameters["a"] a6 = curve.parameters["b"] return a1, a2, a3, a4, a6 else: raise NotImplementedError @public def b_invariants(curve: EllipticCurve) -> Tuple[Mod, ...]: """ Compute the b-invariants of the curve. :param curve: The elliptic curve (only ShortWeierstrass model). :return: A tuple of 4 b-invariants (b2, b4, b6, b8). """ if isinstance(curve.model, ShortWeierstrassModel): a1, a2, a3, a4, a6 = a_invariants(curve) return ( a1 * a1 + 4 * a2, a1 * a3 + 2 * a4, a3**2 + 4 * a6, a1**2 * a6 + 4 * a2 * a6 - a1 * a3 * a4 + a2 * a3**2 - a4**2, ) else: raise NotImplementedError def divpoly0(curve: EllipticCurve, *ns: int) -> Mapping[int, Poly]: """ Basically sagemath's division_polynomial_0 but more clever memory management. As sagemath says: Return the `n^{th}` torsion (division) polynomial, without the 2-torsion factor if `n` is even, as a polynomial in `x`. These are the polynomials `g_n` defined in [MT1991]_, but with the sign flipped for even `n`, so that the leading coefficient is always positive. :param curve: The elliptic curve. :param ns: The values to compute the polynomial for. :return: """ xs = symbols("x") K = FF(curve.prime) Kx = lambda r: Poly(r, xs, domain=K) # noqa x = Kx(xs) b2, b4, b6, b8 = map(lambda b: Kx(int(b)), b_invariants(curve)) ls, _ = dep_map(*ns) mem: Dict[int, Poly] = {} for i, keep in ls: if i == -2: val = mem[-1] ** 2 elif i == -1: val = Kx(4) * x**3 + b2 * x**2 + Kx(2) * b4 * x + b6 elif i == 0: val = Kx(0) elif i < 0: raise ValueError("n must be a positive integer (or -1 or -2)") elif i in (1, 2): val = Kx(1) elif i == 3: val = Kx(3) * x**4 + b2 * x**3 + Kx(3) * b4 * x**2 + Kx(3) * b6 * x + b8 elif i == 4: val = -mem[-2] + (Kx(6) * x**2 + b2 * x + b4) * mem[3] elif i % 2 == 0: m = (i - 2) // 2 val = mem[m + 1] * (mem[m + 3] * mem[m] ** 2 - mem[m - 1] * mem[m + 2] ** 2) else: m = (i - 1) // 2 if m % 2 == 0: val = mem[-2] * mem[m + 2] * mem[m] ** 3 - mem[m - 1] * mem[m + 1] ** 3 else: val = mem[m + 2] * mem[m] ** 3 - mem[-2] * mem[m - 1] * mem[m + 1] ** 3 for dl in set(mem.keys()).difference(keep).difference(ns): del mem[dl] mem[i] = val return mem @public def divpoly(curve: EllipticCurve, n: int, two_torsion_multiplicity: int = 2) -> Poly: """ Compute the n-th division polynomial. :param curve: Curve to compute on. :param n: Scalar. :param two_torsion_multiplicity: Same as sagemath. :return: The division polynomial. """ f: Poly = divpoly0(curve, n)[n] a1, a2, a3, a4, a6 = a_invariants(curve) xs, ys = symbols("x y") x = Poly(xs, xs, domain=f.domain) y = Poly(ys, ys, domain=f.domain) if two_torsion_multiplicity == 0: return f elif two_torsion_multiplicity == 1: if n % 2 == 0: Kxy = lambda r: Poly(r, xs, ys, domain=f.domain) # noqa return Kxy(f) * (Kxy(2) * y + Kxy(a1) * x + Kxy(a3)) else: return f elif two_torsion_multiplicity == 2: if n % 2 == 0: return f * divpoly0(curve, -1)[-1] else: return f else: raise ValueError def mult_by_n_own(curve: EllipticCurve, n: int) -> Tuple[Poly, Poly]: xs, ys = symbols("x y") K = FF(curve.prime) x = Poly(xs, xs, domain=K) Kxy = lambda r: Poly(r, xs, ys, domain=K) # noqa if n == 1: return x, Kxy(1) polys = divpoly0(curve, -2, -1, n - 1, n, n + 1, n + 2) # TODO: All of these fractions may benefit from using # sympy.cancel to get rid of common factors in the numerator and denominator. # Though for large polynomials that might be too much. mx_denom = polys[n] ** 2 if n % 2 == 0: mx_num = x * polys[-1] * polys[n] ** 2 - polys[n - 1] * polys[n + 1] mx_denom *= polys[-1] else: mx_num = x * polys[n] ** 2 - polys[-1] * polys[n - 1] * polys[n + 1] # Alternative that makes the denominator monic by dividing the # numerator by the leading coefficient. Sage does this # simplification when asking for multiplication_by_m with the # x-only=True, as then the poly is an univariate object. # > # > lc = K(mx_denom.LC()) # > mx = (mx_num.quo(lc), mx_denom.monic()) mx = (mx_num, mx_denom) return mx if has_pari: def mult_by_n_pari(curve: EllipticCurve, n: int): pari = cypari2.Pari() # Magic heuristic, plus some constant term for very small polys stacksize = 2 * (n**2 * (40 * curve.prime.bit_length())) + 1000000 stacksizemax = 15 * stacksize pari.default("debugmem", 0) # silence stack warnings pari.allocatemem(stacksize, stacksizemax, silent=True) p = pari(curve.prime) a = pari.Mod(curve.parameters["a"], p) b = pari.Mod(curve.parameters["b"], p) E = pari.ellinit([a, b]) while True: try: mx = pari.ellxn(E, n) break except cypari2.PariError as e: if e.errnum() == 17: # out of stack memory pari.allocatemem(0) else: raise e x = symbols("x") K = FF(curve.prime) mx_num = Poly([int(coeff) for coeff in reversed(mx[0])], x, domain=K) mx_denom = Poly([int(coeff) for coeff in reversed(mx[1])], x, domain=K) return mx_num, mx_denom @public def mult_by_n( curve: EllipticCurve, n: int, x_only: bool = False, use_pari: bool = True ) -> Tuple[Tuple[Poly, Poly], Optional[Tuple[Poly, Poly]]]: """ Compute the multiplication-by-n map on an elliptic curve. :param curve: Curve to compute on. :param n: Scalar. :param x_only: Whether to skip the my computation. :param use_pari: Whether to use the Pari version. :return: A tuple (mx, my) where each is a tuple (numerator, denominator). """ if use_pari and has_pari: mx = mult_by_n_pari(curve, n) else: if use_pari: warnings.warn( "Falling-back to slow mult-by-n map computation due to missing [pari] (cypari2 and libpari) dependency." ) mx = mult_by_n_own(curve, n) if x_only: return mx, None xs, ys = symbols("x y") K = FF(curve.prime) x = Poly(xs, xs, domain=K) y = Poly(ys, ys, domain=K) Kxy = lambda r: Poly(r, xs, ys, domain=K) # noqa a1, a2, a3, a4, a6 = a_invariants(curve) # The following lines compute # my = ((2*y+a1*x+a3)*mx.derivative(x)/m - a1*mx-a3)/2 # just as sage does, but using sympy and step-by-step # tracking the numerator and denominator of the fraction. # > mx.derivative() mxd_num = mx[1] * mx[0].diff() - mx[0] * mx[1].diff() mxd_denom = mx[1] ** 2 # > mx.derivative()/m mxd_dn_num = mxd_num mxd_dn_denom = mxd_denom * Kxy(n) # > (2*y+a1*x+a3)*mx.derivative(x)/m mxd_full_num = mxd_dn_num * (Kxy(2) * y + Kxy(a1) * x + Kxy(a3)) mxd_full_denom = mxd_dn_denom # > a1*mx a1mx_num = Kxy(a1) * mx[0] a1mx_denom = mx[1] # noqa # > a3 a3_num = Kxy(a3) * mx[1] a3_denom = mx[1] # noqa # The mx.derivative part has a different denominator, basically mx[1]^2 * m # so the rest needs to be multiplied by this factor when subtracting. mxd_fact = mx[1] * n my_num = mxd_full_num - a1mx_num * mxd_fact - a3_num * mxd_fact my_denom = mxd_full_denom * Kxy(2) my = (my_num, my_denom) return mx, my