"""Sandbox (non-test points only): finite-t* bias of P(+) against Phi(kappa x), and the linear-theory '+' direction.
Grid: theta in {0.004,0.007,0.012,0.025,0.04}, N in {1000,3000}; mu in {1.5,2,2.5}; t* thresholds 0.05 and 0.1."""
import numpy as np
from math import erf, sqrt
Phi = np.vectorize(lambda z: 0.5 * (1 + erf(z / sqrt(2))))
def phi_lin(mu):
    lam = np.sqrt(mu - 1); al = 1 - mu / 2
    return np.angle(-(1 + 1j * lam) / (lam - 1j * al))
def rhs(a, b, B, mu):
    s = 0.5 * mu * (a + b); return np.cross(B + s, a), np.cross(-B + s, b)
def run(theta, N, mu, thr, M=20000, dt=0.01, seed=7):
    rng = np.random.default_rng(seed); sg = np.sqrt(2.0 / N); eps = np.sin(2 * theta)
    B = np.array([eps, 0.0, np.cos(2 * theta)])
    a = np.zeros((M + 1, 3)); b = np.zeros((M + 1, 3)); a[:, 2] = 1; b[:, 2] = -1
    a[1:, :2] = rng.normal(0, sg, (M, 2)); b[1:, :2] = rng.normal(0, sg, (M, 2))
    out = {}; t = 0.0; todo = sorted(thr)
    while todo:
        k1 = rhs(a, b, B, mu); k2 = rhs(a + dt/2*k1[0], b + dt/2*k1[1], B, mu)
        k3 = rhs(a + dt/2*k2[0], b + dt/2*k2[1], B, mu); k4 = rhs(a + dt*k3[0], b + dt*k3[1], B, mu)
        a = a + dt/6*(k1[0]+2*k2[0]+2*k3[0]+k4[0]); b = b + dt/6*(k1[1]+2*k2[1]+2*k3[1]+k4[1]); t += dt
        du = a[:, 0] + 1j * a[:, 1] - eps
        m2 = np.mean(np.abs(du[1:])**2)
        while todo and m2 >= todo[0]:
            th = todo.pop(0); n = np.exp(1j * phi_lin(mu))
            out[th] = (t, np.mean(np.real(du[1:] * np.conj(n)) > 0), np.degrees(np.angle(du[0] * np.conj(n))))
    return out
rows = []
for theta in (0.004, 0.007, 0.012, 0.025, 0.04):
    for N in (1000, 3000):
        x = np.sin(2*theta) * np.sqrt(N/2)
        for mu in (1.5, 2.0, 2.5):
            r = run(theta, N, mu, (0.05, 0.1)); k = np.sqrt(2/mu)
            for th, (t, P, dang) in r.items():
                rows.append((theta, N, mu, th, x, t, P, float(Phi(k*x)), float(Phi(x)), dang))
print("theta     N    mu  thr     x     t*     P_TWA   Phi(kx)  bias    Phi(x)  noiseless-vs-linear angle")
for r in rows:
    print(f"{r[0]:<6} {r[1]:5d} {r[2]:4} {r[3]:<4} {r[4]:6.3f} {r[5]:5.2f}  {r[6]:.4f}  {r[7]:.4f}  {r[6]-r[7]:+.4f}  {r[8]:.4f}  {r[9]:+6.2f}")
for th in (0.05, 0.1):
    b = np.array([r[6]-r[7] for r in rows if r[3] == th])
    print(f"thr {th}: bias mean {b.mean():+.4f}, max |bias| {np.abs(b).max():.4f}, rms {np.sqrt((b**2).mean()):.4f}  (sampling sd <= 0.0035)")
