File size: 4,624 Bytes
848d4b7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
143
144
145
#!/usr/bin/env python3
"""
Validator for problem 040: Signed Autocorrelation Constant C' Upper Bound

The signed autocorrelation constant C' is defined as:
    C' = inf_f max_t (f*f)(t) / (∫f)^2
where f may be positive or negative (not restricted to be non-negative)
and is supported on [-1/4, 1/4].

The current best upper bound is C' ≤ 1.4557 (AlphaEvolve, 2025).

The model provides a step function as a list of real values on N equal-width
subintervals of [-1/4, 1/4]. The validator computes the autoconvolution ratio
and checks if it improves the best known upper bound.

Expected input format:
    {"values": [v_0, v_1, ..., v_{N-1}]}
    or [v_0, v_1, ..., v_{N-1}]
"""

import argparse
from typing import Any

import numpy as np
from scipy.signal import fftconvolve

from . import ValidationResult, load_solution, output_result, success, failure


BEST_KNOWN_UPPER = 1.4557
MIN_INTERVALS = 10
MAX_INTERVALS = 1_000_000


def compute_autoconvolution_ratio(values: np.ndarray) -> float:
    """
    Compute max_t (f*f)(t) / (∫f)^2 for a step function.

    The function f is defined on N equal-width subintervals of [-1/4, 1/4].
    Each subinterval has width h = (1/2) / N.

    The autoconvolution (f*f)(t) = ∫ f(t-x)f(x) dx is computed via
    discrete convolution of the step function values scaled by the
    subinterval width h.

    This is equivalent to the AlphaEvolve evaluator formula:
        score = 2n * max(convolve(a, a)) / (sum(a))^2
    since h = 1/(2n), so max(conv)*h / (sum(a)*h)^2
         = max(conv) / (sum(a)^2 * h) = 2n * max(conv) / sum(a)^2.
    """
    n = len(values)
    h = 0.5 / n  # width of each subinterval

    # Discrete convolution: (f*f) sampled at points spaced by h
    # fftconvolve gives the convolution of the coefficient sequences;
    # multiply by h to account for the integral approximation
    conv = fftconvolve(values, values) * h

    max_conv = np.max(conv)
    integral_f = np.sum(values) * h

    if integral_f == 0:
        return float('inf')

    return max_conv / (integral_f ** 2)


def validate(solution: Any) -> ValidationResult:
    """
    Validate a signed autocorrelation upper bound construction.

    Args:
        solution: Dict with 'values' key or list of real values

    Returns:
        ValidationResult with autoconvolution ratio
    """
    try:
        if isinstance(solution, dict) and 'values' in solution:
            values_data = solution['values']
        elif isinstance(solution, list):
            values_data = solution
        else:
            return failure("Invalid format: expected dict with 'values' or list")

        values = np.array(values_data, dtype=np.float64)
    except (ValueError, TypeError) as e:
        return failure(f"Failed to parse values: {e}")

    if values.ndim != 1:
        return failure(f"Values must be a 1D array, got {values.ndim}D")

    n = len(values)
    if n < MIN_INTERVALS:
        return failure(f"Need at least {MIN_INTERVALS} intervals, got {n}")

    if n > MAX_INTERVALS:
        return failure(f"Too many intervals ({n}), maximum is {MAX_INTERVALS}")

    # Check all entries are finite reals (reject NaN/inf)
    if not np.all(np.isfinite(values)):
        return failure("All values must be finite real numbers (no NaN or inf)")

    # Check function is not identically zero
    if np.all(values == 0):
        return failure("Function is identically zero")

    # Check sum is nonzero (otherwise the ratio is undefined)
    if np.sum(values) == 0:
        return failure("Sum of values is zero (autoconvolution ratio is undefined)")

    ratio = compute_autoconvolution_ratio(values)

    if not np.isfinite(ratio):
        return failure(
            "Computed ratio is not finite, indicating a numerical issue",
            autoconvolution_ratio=float(ratio)
        )

    return success(
        f"Step function with {n} intervals achieves autoconvolution ratio {ratio:.6f} "
        f"(best known: {BEST_KNOWN_UPPER})",
        num_intervals=n,
        autoconvolution_ratio=ratio,
        best_known_upper=BEST_KNOWN_UPPER,
        improves_bound=ratio < BEST_KNOWN_UPPER
    )


def main():
    parser = argparse.ArgumentParser(
        description='Validate signed autocorrelation upper bound construction'
    )
    parser.add_argument('solution', help='Solution as JSON string or path to JSON file')
    parser.add_argument('--verbose', '-v', action='store_true', help='Verbose output')
    args = parser.parse_args()

    solution = load_solution(args.solution)
    result = validate(solution)
    output_result(result)


if __name__ == '__main__':
    main()