# [project] Block 4: toy mollifier. M(s) = sum_{n<=y} mu(n) P(log(y/n)/log y) n^{-s}, P(x) = x (Levinson's shape).
# Compare log|zeta| with log|zeta*M| on t in [1000, 1100] (step 0.01) at sigma = 1/2 and at Levinson's edge
# sigma0 = 1/2 - R/log T (R = 1.3, T = 1050). Double-computed:
#   A = mpmath.zeta + mollifier summed in mpmath;  C = em_numpy (own Euler–Maclaurin) + mollifier in numpy.
# Small finite-height illustration only — not a theorem, nothing asymptotic. Output: toy_mollifier.json
import json, numpy as np, mpmath as mp
import em_numpy as E
mp.mp.dps = 20
T0, T1, STEP, R = 1000.0, 1100.0, 0.01, 1.3
t = np.round(np.arange(T0, T1 + 1e-9, STEP), 6)
SIG0 = 0.5 - R/np.log(0.5*(T0 + T1))
YS = [1, 5, 10, 20, 32, 60]               # y = 32 ~ T^(1/2); y = 1 means "no mollifier" (M = 1)
def mobius(n):
    m, p, k = n, 2, 0
    while p*p <= m:
        if m % p == 0:
            m //= p
            if m % p == 0: return 0
            k += 1
        p += 1
    if m > 1: k += 1
    return (-1)**k
def coeffs(y):
    if y == 1: return {1: 1.0}
    return {n: mobius(n)*np.log(y/n)/np.log(y) for n in range(1, y + 1) if mobius(n) != 0 and n < y}
def stats(v):
    a = np.abs(v); la = np.log(a)
    return {"mean_log_abs": float(la.mean()), "var_log_abs": float(la.var()), "mean_abs2": float((a**2).mean()),
            "median_abs": float(np.median(a)), "frac_abs_in_[1/2,2]": float(np.mean((a >= 0.5) & (a <= 2))),
            "max_abs": float(a.max())}
out = {"what": __doc__ if __doc__ else "toy mollifier", "t_range": [T0, T1], "step": STEP, "sigma0": float(SIG0),
       "P": "x", "y_values": YS, "A_mpmath": {}, "C_numpy": {}, "max_abs_diff_zetaM_A_vs_C": {}}
for sig in (0.5, float(SIG0)):
    zA = np.array([complex(mp.zeta(mp.mpc(sig, x))) for x in t])
    zC = E.zeta(sig, t)
    key = f"sigma={sig:.6f}"
    out["A_mpmath"][key] = {}; out["C_numpy"][key] = {}; out["max_abs_diff_zetaM_A_vs_C"][key] = {}
    for y in YS:
        c = coeffs(y)
        MA = np.array([complex(mp.fsum(mp.mpf(a)*mp.power(n, -mp.mpc(sig, x)) for n, a in c.items())) for x in t])
        ns = np.array(list(c.keys()), float); cs = np.array(list(c.values()))
        MC = np.exp(-np.outer(sig + 1j*t, np.log(ns))) @ cs
        fA, fC = zA*MA, zC*MC
        out["A_mpmath"][key][str(y)] = stats(fA); out["C_numpy"][key][str(y)] = stats(fC)
        out["max_abs_diff_zetaM_A_vs_C"][key][str(y)] = float(np.max(np.abs(fA - fC)))
        s = out["A_mpmath"][key][str(y)]
        print(key, "y=%3d" % y, " ".join(f"{k}={v:.4f}" for k, v in s.items()), "absAC=%.1e" % out["max_abs_diff_zetaM_A_vs_C"][key][str(y)], flush=True)
out["what"] = "toy mollifier, P(x)=x, A=mpmath vs C=em_numpy; y=1 is no mollifier"
out["max_abs_stat_diff_A_vs_C"] = max(abs(out["A_mpmath"][k][y][m] - out["C_numpy"][k][y][m])
    for k in out["A_mpmath"] for y in out["A_mpmath"][k] for m in out["A_mpmath"][k][y])
print("max |stat_A - stat_C| =", out["max_abs_stat_diff_A_vs_C"])
json.dump(out, open("toy_mollifier.json", "w"), indent=1)
