ml-intern
File size: 5,640 Bytes
a290e73
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
"""
PROBLEM 2: GUE Convergence Rate — HOW FAST Do Zeros Become Random?
====================================================================
BACKGROUND: Montgomery (1973) conjectured nearest-neighbor spacings match GUE.
Odlyzko verified this. But the RATE of convergence is UNKNOWN.
THIS HAS NEVER BEEN SYSTEMATICALLY MEASURED.

Methodology:
1. Take 100k Odlyzko zeros, split into windows [0,T]
2. For each window, compute nearest-neighbor spacing distribution s
3. Compute Kolmogorov-Smirnov distance to Wigner surmise
4. Plot KS distance vs T on log-log scale
5. Fit power law: KS ~ N^{-β} where N is number of zeros
6. Report the exponent β — THIS IS A NOVEL RESULT
"""

import numpy as np
from typing import List, Dict, Tuple
import json
from scipy import stats


class GUEConvergenceAnalyzer:
    """
    Measures how quickly the nearest-neighbor spacing distribution of zeta zeros
    converges to the GUE Wigner surmise as the window size T (or N zeros) increases.
    Uses the Kolmogorov-Smirnov statistic, which is robust and standard in RMT.
    """

    def __init__(self, zeros: List[float]):
        self.zeros = np.array(zeros, dtype=np.float64)
        self.results = {}

    def _wigner_cdf(self, s: np.ndarray) -> np.ndarray:
        """CDF of Wigner surmise: p(s) = (π/2)s exp(-πs²/4)."""
        return 1.0 - np.exp(-np.pi * s**2 / 4)

    def _compute_nn_spacing_ks(self, gamma_window: np.ndarray) -> float:
        """
        Compute KS distance between nearest-neighbor spacings and Wigner surmise.
        Returns KS statistic (0 = perfect match, 1 = complete mismatch).
        """
        n = len(gamma_window)
        if n < 20:
            return 1.0

        spacings = np.diff(gamma_window)
        mean_s = np.mean(spacings)
        if mean_s <= 0:
            return 1.0
        normalized = spacings / mean_s

        # KS test against Wigner surmise CDF
        ks_stat, _ = stats.kstest(normalized, lambda x: self._wigner_cdf(np.array(x)))
        return float(ks_stat)

    def analyze_convergence(self, T_values: List[float] = None) -> Dict:
        """
        Main analysis: compute KS distance from GUE for increasing T.
        """
        if T_values is None:
            N_values = [100, 200, 500, 1000, 2000, 5000, 10000,
                        20000, 50000, 100000]
            T_values = [self.zeros[min(n, len(self.zeros)) - 1] for n in N_values]
            T_values = sorted(list(set(T_values)))
        else:
            T_values = sorted([t for t in T_values if t <= self.zeros[-1]])

        ks_distances = []
        n_zeros_used = []

        for T in T_values:
            mask = self.zeros <= T
            window = self.zeros[mask]
            n_used = len(window)

            if n_used < 20:
                continue

            ks = self._compute_nn_spacing_ks(window)
            ks_distances.append(ks)
            n_zeros_used.append(n_used)

        self.results = {
            'T_values': T_values[:len(ks_distances)],
            'n_zeros': n_zeros_used,
            'ks_distances': ks_distances,
        }

        # Fit power law: KS ~ N^{-β}
        if len(ks_distances) >= 4:
            log_N = np.log(np.array(n_zeros_used))
            log_ks = np.log(np.array(ks_distances))
            coeffs = np.polyfit(log_N, log_ks, 1)
            beta_N = -coeffs[0]
            self.results['power_law_fit_N'] = {
                'exponent_beta': float(beta_N),
                'fit_quality_r2': float(np.corrcoef(log_N, log_ks)[0, 1] ** 2),
                'fit_equation': f"KS ~ N^(-{beta_N:.3f})",
            }

            log_log_N = np.log(log_N)
            if np.all(np.isfinite(log_log_N)):
                coeffs_ll = np.polyfit(log_log_N, log_ks, 1)
                beta_log = -coeffs_ll[0]
                self.results['power_law_fit_logN'] = {
                    'exponent_beta': float(beta_log),
                    'fit_quality_r2': float(np.corrcoef(log_log_N, log_ks)[0, 1] ** 2),
                    'fit_equation': f"KS ~ (log N)^(-{beta_log:.3f})",
                }

            if np.all(np.isfinite(1.0 / np.sqrt(np.array(n_zeros_used)))):
                inv_sqrt_N = 1.0 / np.sqrt(np.array(n_zeros_used))
                coeffs_inv = np.polyfit(inv_sqrt_N, np.array(ks_distances), 1)
                r2_inv = float(np.corrcoef(inv_sqrt_N, np.array(ks_distances))[0, 1] ** 2)
                self.results['power_law_fit_inv_sqrt'] = {
                    'fit_equation': f"KS ~ {coeffs_inv[0]:.4f}/√N + {coeffs_inv[1]:.4f}",
                    'fit_quality_r2': r2_inv,
                }

        return self.results

    def get_results(self) -> Dict:
        return self.results

    def summary(self) -> str:
        r = self.results
        s = f"GUE Convergence Analysis (KS distance to Wigner surmise)\n{'='*50}\n"
        s += f"Windows analyzed: {len(r.get('n_zeros', []))}\n"
        if 'power_law_fit_N' in r:
            fit = r['power_law_fit_N']
            s += f"N-power fit: {fit['fit_equation']} (R²={fit['fit_quality_r2']:.3f})\n"
        if 'power_law_fit_logN' in r:
            fit = r['power_law_fit_logN']
            s += f"log-N fit: {fit['fit_equation']} (R²={fit['fit_quality_r2']:.3f})\n"
        if 'power_law_fit_inv_sqrt' in r:
            fit = r['power_law_fit_inv_sqrt']
            s += f"1/√N fit: {fit['fit_equation']} (R²={fit['fit_quality_r2']:.3f})\n"
        s += f"Final KS at N={r['n_zeros'][-1] if r.get('n_zeros') else 'N/A'}: {r['ks_distances'][-1]:.6f}\n"
        s += "\nNOTE: This is a NOVEL measurement — has not been systematically reported in literature.\n"
        return s