generate.py

back to table · edit · history · where entries came from · files · download

9805 bytes, as of the version from 2026-09-16 12:57 (current). Recorded here, not run.

"""Core thresholds of random $k$-uniform hypergraphs -- numberdb.org/T273.

For k >= 3 and r >= 2, c_{k,r} is the edge density m/n at which the random
k-uniform hypergraph with n vertices and m edges acquires a nonempty r-core.
Molloy and Cain-Wormald give

    c_{k,r} = min over lambda > 0 of
              lambda / (k * P(Poisson(lambda) >= r - 1)^(k - 1)).

This table stores c_{k,r} for 3 <= k <= 8 and 2 <= r <= 8 at 100 digits in
ball arithmetic.

Run it with SageMath:

    $ sage -pip install numberdb          # once
    $ sage -python generate.py            # check the table against this code
    $ sage -python generate.py --publish  # fill the draft, with NUMBERDB_API_KEY set

The minimiser lambda_{k,r} is enclosed by bisection on the sign of

    g(lambda) = Q(lambda,r-1) - (k-1) exp(-lambda) lambda^(r-1)/(r-2)!,

where Q(lambda,s) = P(Poisson(lambda) >= s). The signs at both ends of the
final bracket are checked in ball arithmetic. Evaluating c_{k,r} on that
bracket gives the stored enclosure.
"""

import os
import sys
from decimal import Decimal

import numberdb.sage as numberdb
from numberdb._write import to_text
from sage.rings.integer_ring import ZZ
from sage.rings.real_arb import RealBallField
from sage.rings.real_mpfr import RealField


TABLE = os.environ.get("NUMBERDB_TABLE", "T273")

# Measured before filling the draft: 42 entries, longest value 119 characters,
# and an entries block of about 13 KB including comments.
K_MIN = 3
K_MAX = 8
R_MIN = 2
R_MAX = 8

# Bits of working precision beyond what the written digits need. Measured at
# 100 digits: every result ball has radius below 9e-106.
WORKING_GUARD = 64

# Decimal places of the bracket around the minimiser beyond the digits written.
# On this rectangle, the value enclosure is still under 9e-106 in radius.
BRACKET_GUARD = 6

COMMENT_DIGITS = 12

PUBLISHED_TEN_DIGITS = {
    (3, 2): "0.8184691607",
    (3, 3): "1.5528299396",
    (4, 2): "0.7722798398",
    (5, 2): "0.7017802665",
}


def _key_from_stdin():
    if os.environ.get("NUMBERDB_KEY_FROM_STDIN") != "1":
        return
    token = sys.stdin.read().strip()
    if "=" in token and token.split("=", 1)[0].isupper():
        token = token.split("=", 1)[1].strip().strip("'\"")
    if token:
        os.environ["NUMBERDB_API_KEY"] = token


def poisson_tail(x, s):
    """Q(x,s) = P(Poisson(x) >= s), for integer s >= 1 and a ball x."""
    if s < 1:
        raise ValueError("the finite tail formula here expects s >= 1")
    total = x.parent()(0)
    term = x.parent()(1)
    for j in range(s):
        if j:
            term = term * x / j
        total += term
    return 1 - (-x).exp() * total


def minimiser_function(x, k, r):
    """The numerator of the derivative of lambda / Q(lambda,r-1)^(k-1)."""
    return (
        poisson_tail(x, r - 1)
        - (k - 1) * (-x).exp() * x ** (r - 1) / ZZ(r - 2).factorial()
    )


def threshold_from_lambda(x, k, r):
    """The hypergraph r-core threshold at a candidate minimiser."""
    return x / (k * poisson_tail(x, r - 1) ** (k - 1))


def minimiser(k, r, digits):
    """A ball enclosing the unique positive minimiser lambda_{k,r}."""
    if k < 2:
        raise ValueError("the formula is used here only for k >= 2")
    if r < 2:
        raise ValueError("the formula is used here only for r >= 2")
    bits = numberdb.bits(digits, losing=WORKING_GUARD)
    RBF, RR = RealBallField(bits), RealField(bits)
    delta = RR(10) ** (-(digits + BRACKET_GUARD))
    lo = RR(r - 1) - RR(1) / RR(k - 1)
    hi = RR(4 * (k + r) + 10)

    if not (minimiser_function(RBF(lo), k, r) < 0):
        raise ArithmeticError("left endpoint does not have negative sign for k=%s, r=%s" % (k, r))
    while not (minimiser_function(RBF(hi), k, r) > 0):
        hi *= 2
        if hi > 10000:
            raise ArithmeticError("right endpoint did not bracket the minimiser for k=%s, r=%s" % (k, r))

    while hi - lo > delta / 4:
        mid = (lo + hi) / 2
        value = minimiser_function(RBF(mid), k, r)
        if value.contains_zero():
            raise ArithmeticError(
                "could not decide minimiser sign at lambda=%s for k=%s, r=%s" % (mid, k, r)
            )
        if value < 0:
            lo = mid
        else:
            hi = mid

    x0 = (lo + hi) / 2
    left = minimiser_function(RBF(x0 - delta), k, r)
    right = minimiser_function(RBF(x0 + delta), k, r)
    if not (left < 0 and right > 0):
        raise ArithmeticError("no sign change across minimiser bracket for k=%s, r=%s" % (k, r))
    return RBF(x0).add_error(delta)


def threshold(k, r, digits):
    """(c_{k,r}, lambda_{k,r}) as real balls."""
    lam = minimiser(k, r, digits)
    value = threshold_from_lambda(lam, k, r)
    if not (0 < value and value < 100):
        raise ArithmeticError("threshold came out as %s at lambda=%s for k=%s, r=%s" % (value, lam, k, r))
    return value, lam


class HypergraphCoreThresholds(numberdb.Generator):
    """Generator for T273, the table of hypergraph core thresholds."""

    table = TABLE
    parameters = ("k", "r")
    type = "R"
    digits = 100
    rigour = "proven"

    def enumerate(self):
        for k in range(K_MIN, K_MAX + 1):
            for r in range(R_MIN, R_MAX + 1):
                yield {"k": str(k), "r": str(r)}

    def value(self, params, digits):
        k = int(params["k"])
        r = int(params["r"])
        value, lam = threshold(k, r, digits)
        comment = r"The minimum is attained at $\lambda_{%d,%d}=%s$." % (
            k,
            r,
            to_text(lam, COMMENT_DIGITS),
        )
        return {"number": value, "comment": comment}


def _mpmath_threshold(k, r, digits=120):
    """Independent numerical value using mpmath's incomplete gamma function."""
    from mpmath import mp

    mp.dps = digits + 30
    factorial = mp.factorial(r - 2)

    def q(x):
        return mp.gammainc(r - 1, 0, x, regularized=True)

    def g(x):
        return q(x) - (k - 1) * mp.e ** (-x) * x ** (r - 1) / factorial

    left = mp.mpf(r - 1) - mp.mpf(1) / mp.mpf(k - 1)
    right = mp.mpf(4 * (k + r) + 10)
    while g(right) <= 0:
        right *= 2
    for _ in range(4 * digits):
        mid = (left + right) / 2
        if g(mid) < 0:
            left = mid
        else:
            right = mid
    root = (left + right) / 2
    return root / (k * q(root) ** (k - 1))


def run_integrity_checks():
    """Checks that do not share the generator's finite-sum implementation."""
    bits = numberdb.bits(120, losing=WORKING_GUARD)
    RBF = RealBallField(bits)
    tolerance = RBF("1e-95")

    for k in range(K_MIN, K_MAX + 1):
        for r in range(R_MIN, R_MAX + 1):
            value, _ = threshold(k, r, 100)
            independent_value = _mpmath_threshold(k, r)
            independent = RBF(str(independent_value))
            if not (abs(value - independent) < tolerance):
                raise ArithmeticError("mpmath comparison failed for k=%d, r=%d" % (k, r))
            expected = PUBLISHED_TEN_DIGITS.get((k, r))
            if expected is not None:
                places = len(expected.split(".")[1])
                independent_text = str(independent_value)
                rounded = Decimal(independent_text).quantize(Decimal("1e-%d" % places))
            if expected is not None and not (independent_text.startswith(expected) or str(rounded) == expected):
                raise ArithmeticError("published ten-digit check failed for k=%d, r=%d" % (k, r))

    table = numberdb.table("T156")
    graph_values = table["Numbers"]
    for r in range(3, R_MAX + 1):
        value, _ = threshold(2, r, 100)
        stored_half = RBF(graph_values[str(r)]["number"]) / 2
        if not (abs(value - stored_half) < tolerance):
            raise ArithmeticError("T156 half-value comparison failed for r=%d" % r)


def fill_draft_once(generator, message):
    """Fill a fresh draft without the client's empty upsert probe."""
    from numberdb._generate import (
        _check_precision,
        _check_rigour,
        _producer,
        _run_name,
        _source_files,
    )
    from numberdb._write import Entries, attach, submit_entries, to_text

    table = generator.table
    run = _run_name(generator)
    entries = Entries(*generator.parameters)

    for params in generator.enumerate():
        params = dict(params)
        wanted = generator.digits_for(params)
        entry = generator._entry(params, wanted)
        value = entry["number"]
        identity = ",".join(str(params[name]) for name in generator.parameters)
        _check_rigour(generator, table, identity, value)

        written = to_text(value, wanted, generator.format)
        _check_precision(table, identity, written, wanted, lowering=False)

        record = dict(entry)
        record.pop("digits", None)
        entries.add(**params, **record, digits=wanted)

    answer = submit_entries(
        table,
        entries,
        message=message,
        produced_by=_producer(generator, os.environ.get("NUMBERDB_ASSISTED_BY", "")),
        upsert=False,
        run=run,
        rigour=generator.rigour,
    )

    for name, body in sorted(_source_files(generator).items()):
        attach(table, name, body, run=run, message=message, rigour=generator.rigour)

    return answer


if __name__ == "__main__":
    _key_from_stdin()
    generator = HypergraphCoreThresholds()
    run_integrity_checks()
    if "--publish" in sys.argv or os.environ.get("NUMBERDB_PUBLISH") == "1":
        print(fill_draft_once(
            generator,
            message=(
                "random k-uniform hypergraph r-core thresholds c_{k,r} for "
                "3 <= k <= 8 and 2 <= r <= 8"
            ),
        ))
    else:
        report = generator.verify(sample=None)
        print(report)
        sys.exit(0 if report.ok else 1)