Datasets:
Tasks:
Other
Formats:
json
Languages:
English
Size:
1K - 10K
ArXiv:
Tags:
mathematics
lorentzian-polynomials
matroids
algorithmic-probability
approximate-exchange
exact-arithmetic
License:
File size: 19,953 Bytes
66937d8 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 | #!/usr/bin/env python3
"""Generate the deterministic, exact research-record collection.
No external dependencies. This constructs analytic records; it does not run
an empirical benchmark or establish the manuscript's all-rank theorems.
The single 'train' split is a Hub collection partition, not a claim that this
is a training corpus or that an evaluation protocol exists.
"""
from __future__ import annotations
import argparse
from fractions import Fraction
import hashlib
import itertools
import json
from math import comb, isqrt
from pathlib import Path
VERSION = "1.0.0"
SOURCE_PDF = "manuscript/Elyxune_Manuscript.pdf"
SOURCE_TEX = "manuscript/Elyxune_Manuscript.tex"
STATUS = "proved_in_manuscript"
PRIORITY = "not_exhaustively_established"
def fraction_text(x: Fraction) -> str:
return f"{x.numerator}/{x.denominator}"
def family_weight(rank: int, s: Fraction, lam: Fraction, basis: tuple[int, ...]) -> Fraction:
"""Ordinary multi-affine coefficient mu(C), Eq. (9), one-based labels."""
j = sum(i <= rank for i in basis)
h = (s * lam) ** (j * (j - 1) // 2)
return (j + 1) * s**j * h / comb(rank, j) if 2*rank in basis else h / comb(rank, j)
def claim_records(source_hash: str) -> list[dict]:
bh = "https://web.math.princeton.edu/~huh/Lorentzian.pdf"
algvv = "https://arxiv.org/pdf/2004.07220"
bhkl = "https://arxiv.org/abs/2607.15375"
items = [
("sharp_exchange_chain", 1, "theorem", "main", "Sharp root, sum and maximum exchange",
"P <= (2^(d-1)/d)*(sum_j sqrt(E_j))^2 <= 2^(d-1)*sum_j E_j <= d*2^(d-1)*max_j E_j.",
["The generating polynomial is homogeneous, multi-affine and Lorentzian with nonnegative coefficients.", "S,T have positive weights; i is prescribed in S\\T; d=|S\\T|; E_j=mu(S-i+j)*mu(T+i-j)."],
"Contract the common intersection, aggregate variables with arbitrary positive weights into a ternary polynomial, telescope quadratic Hessian determinants, and optimize the weights. An explicit interior family proves optimality.",
["Optimality is a supremum statement; attainment by one polynomial is not asserted.", "The input class and prescribed-element quantifier are essential."], [bh]),
("exact_max_exchange_constant", 2, "corollary", "constant", "Exact universal approximate-exchange constant",
"C_r = r*2^(r-1); the exact constant at exchange distance d is d*2^(d-1).",
["C_r is universal over all ground sets, supports, positive-weight pairs and prescribed differing elements in the rank-r homogeneous multi-affine log-concave class."],
"The sharp chain gives the upper bound; equal-exchange rational families approach it with exchange distance r.",
["This is an intrinsic inequality constant, not a lower bound for an optimization algorithm."], [algvv,bh]),
("sharp_lq_exchange_profile", 3, "theorem", "lq", "Complete sharp l_q exchange profile",
"P <= K_(d,q)*||E||_q, with K_(d,q)=2^(d-1)*d^(1-1/q), for 1/2<=q<=infinity and 1/infinity=0.",
["Use the unnormalized profile ||E||_q=(sum_j E_j^q)^(1/q), and the maximum at q=infinity.", "The assumptions and exchange products are those of Theorem 1."],
"Apply finite-dimensional Holder to the sharp root-sum bound. Equal exchange products in the construction prove sharpness for every q simultaneously.",
["For q<1 this profile is a quasi-norm.", "No claim is made for q<1/2."], [bh]),
("ternary_telescoping", 4, "lemma", "telescoping", "Ternary telescoping endpoint bound",
"h_d*g_0 <= 2^(d-1)*h_1*g_(d-1) for the normalized nonnegative Lorentzian polynomial F=H(x,y)+u*G(x,y) of degree d.",
["F is affine in u, H has degree d and G has degree d-1.", "Coefficients h_j,g_j use the factorial-normalized monomial convention in the manuscript."],
"The 3x3 quadratic Hessian determinant gives h_(j+2)/h_(j+1)<=2*g_(j+1)/g_j in the positive case. Telescope; handle zeros by binary epsilon mixing and continuity.",
["Nonnegative linear-substitution closure is a standard structural input."], [bh]),
("rational_near_extremizers", 5, "corollary", "rational", "Finite rational near-extremizers",
"For each d>=2 and 0<epsilon<1, positive rational weights with nonsingular active Hessians attain at least (1-epsilon) times every optimal l_q constant simultaneously.",
["Choose integer N>=max(2,(d-1)/(2*epsilon)), lambda=2-1/N and s=1/(2*N)."],
"The ratio to the supremal constant is (1-1/(2*N))^(d-1), bounded below by Bernoulli's inequality. Coefficients have an explicit rational formula.",
["These are constructed mathematical coefficients, not empirical measurements.", "Sharpness is approached; exact equality in the interior is not claimed."], [bh]),
("polynomial_exchange_negative_answer", 6, "corollary", "refutation", "Negative answer to polynomial approximate exchange",
"No universal polynomial p(r) guarantees prescribed-element symmetric exchange for every rank-r homogeneous multi-affine log-concave generating polynomial.",
["The prescribed-element definition is Definition 24 in arXiv:2004.07220v2.", "Fix lambda=3/2, s=1/4, t=3/8 in the geometric family."],
"For S=B+u, T=A and prescribed i=u, every possible exchange has ratio r*(3/2)^(r-1), which exceeds every fixed polynomial for large r.",
["The mathematical implication targets the unnumbered Section 4 conjecture as printed.", "Historical priority is not exhaustively established.", "Weak exchange notions with different quantifiers are separate."], [algvv]),
("ternary_local_criterion", 7, "proposition", "ternarycriterion", "Exact positive ternary local criterion",
"For positive h_j,g_j, F=H+u*G is Lorentzian iff g_j*g_(j+2)<=g_(j+1)^2 and G_j/H_j + H_(j+1)/G_j <=2, with H_j=h_(j+1)/h_j and G_j=g_(j+1)/g_j.",
["F has degree d>=2, is affine in u, and its normalized binary coefficient sequences are strictly positive.", "Index ranges are as in Proposition 7."],
"Exhaust the quadratic derivatives. Those containing u give binary log-concavity; the other 3x3 Hessians give the displayed determinant condition.",
["The positive-coefficient assumption is part of this local characterization."], [bh]),
("ternary_transfer_region", 8, "theorem", "transferregion", "Exact transfer region and realizers",
"For a<b, m=b-a, P=(g_b*h_a)/(g_a*h_b), Q=(h_(b+1)*g_a)/(h_(a+1)*g_b), the exact feasible positive region is P^(1/m)+Q^(1/m)<=2.",
["The positive Lorentzian ternary assumptions of Proposition 7 hold.", "0<=a<b<=d-1."],
"Multiplicative Holder combines local ratios x_j+y_j<=2. Every feasible pair is realized by a geometric seed; equality forces constant local transfer ratios on the interval.",
["The statement characterizes these two ratios, not every coefficient of every Lorentzian polynomial.", "It does not classify all strong hyperfield relations."], [bh]),
("linear_extension_criterion", 9, "theorem", "extension", "Exact linear-extension criterion",
"For f(u,z)=h(z)+u*g(z), r>=2 and M-convex support, Lorentzianity is equivalent to g Lorentzian and A_alpha restricted to ker(v_alpha^T) negative semidefinite when v_alpha!=0; when v_alpha=0, A_alpha has at most one positive eigenvalue.",
["h has degree r, g degree r-1, and coefficients are nonnegative.", "A_alpha=Hessian(partial^alpha h), v_alpha=gradient(partial^alpha g), |alpha|=r-2."],
"A bordered Hessian is congruent to one hyperbolic plane plus the restriction of A_alpha to ker(v_alpha^T). For fixed g the restrictions are linear matrix inequalities in h.",
["The convex description is on a fixed full positive support stratum.", "Arbitrary support changes are not declared convex."], [bh]),
("strong_hyperfield_parameter_lower_bound", None, "derived_consequence", "Hyperfield and phase-completion consequences", "Necessary growing strong hyperfield parameter",
"A parameter representing every Lorentzian coefficient vector on U_(d,2d) strongly over T_p must satisfy p>=1+(d-1)*log(2)/log(d).",
["d>=2; strong relations use polygon closure of p-th roots.", "Use the family relation with products proportional to (d*lambda^(d-1),1,...,1)."],
"The one displayed strong relation requires (d*lambda^(d-1))^(1/p)<=d; let lambda increase to 2.",
["This is necessary, not a sufficient parameter for all coefficients on a fixed support.", "Question 8.6 is excluded."], [bhkl]),
("integer_strong_T2_obstruction", None, "explicit_witness", "Hyperfield and phase-completion consequences", "Six-element integer obstruction",
"The rank-3 integer cubic specified in the manuscript is Lorentzian with active inertia (1,4,0), but its relation products (154,16,16,16) fail strong T_2 because 154>144.",
["A={1,2,3}, B={4,5}, u=6; coefficient types are 14,11,4,8 as specified in the manuscript."],
"Exact characteristic polynomials certify the six active Hessian signatures; the explicit four-product inequality certifies failure.",
["This is a finite mathematical witness, not a benchmark measurement."], []),
("minimality_strong_T2_six_elements", None, "derived_consequence", "Hyperfield and phase-completion consequences", "Minimal ground-set size for strong T_2 failure",
"Six is the smallest ground-set size of a matroid supporting a multi-affine Lorentzian coefficient vector that fails strong T_2 representability.",
["The input is a multi-affine Lorentzian coefficient vector on a matroid."],
"Lorentzianity implies weak T_2. For n<=5 every strong relation has at most three potential terms, so it is trivial or weak. The integer cubic supplies the six-element counterexample.",
["This minimum concerns strong T_2 representability, not actual complex squared-minor completion."], [bhkl]),
]
return [{
"record_id": cid, "title": title, "statement_number": number,
"statement_kind": kind, "source_anchor": anchor,
"statement": statement, "assumptions": assumptions,
"proof_summary": proof, "limitations": limits,
"proof_status": STATUS, "historical_priority": PRIORITY,
"proof_assistant_verified": False,
"source_pdf": SOURCE_PDF, "source_tex": SOURCE_TEX,
"source_tex_sha256": source_hash, "context_references": refs,
"record_provenance": "research_statement_from_included_manuscript",
"dataset_version": VERSION,
} for cid,number,kind,anchor,title,statement,assumptions,proof,limits,refs in items]
def family_records(source_hash: str) -> list[dict]:
rows=[]
for rank in range(2,7):
specs=[
("polynomial_exchange_counterexample", Fraction(1,4), Fraction(3,2), None),
("strict_reference", Fraction(1,10), Fraction(9,5), None),
("near_optimal_N_100r", Fraction(1,200*rank), Fraction(2)-Fraction(1,100*rank), 100*rank),
]
for family_id,s,lam,N in specs:
assert 1 < lam < 2 and 0 < s < 2-lam
t=s*lam
assert 0<t<1
weights={basis:family_weight(rank,s,lam,basis) for basis in itertools.combinations(range(1,2*rank+1),rank)}
A=tuple(range(1,rank+1)); B=set(range(rank+1,2*rank));u=2*rank
target=weights[A]*weights[tuple(sorted(B|{u}))]
expected=rank*lam**(rank-1)
for i in A:
alt=weights[tuple(sorted(set(A)-{i}|{u}))]*weights[tuple(sorted(B|{i}))]
assert target/alt==expected
assert len(weights)==comb(2*rank,rank)
for basis,w in weights.items():
assert w>0
rows.append({
"record_id": f"{family_id}_r{rank}_"+"-".join(map(str,basis)),
"family_id": family_id,"rank":rank,"ground_set_size":2*rank,
"matroid_support":f"U_({rank},{2*rank})","basis":list(basis),
"a_count":sum(i<=rank for i in basis),"u_in_basis":u in basis,
"label_convention":"one_based_A_1_to_r_B_rplus1_to_2rminus1_u_2r",
"s_numerator":str(s.numerator),"s_denominator":str(s.denominator),
"lambda_numerator":str(lam.numerator),"lambda_denominator":str(lam.denominator),
"t_numerator":str(t.numerator),"t_denominator":str(t.denominator),
"near_optimal_N":N,
"coefficient_numerator":str(w.numerator),"coefficient_denominator":str(w.denominator),
"coefficient_exact":fraction_text(w),
"target_to_each_exchange_numerator":str(expected.numerator),
"target_to_each_exchange_denominator":str(expected.denominator),
"target_to_each_exchange_exact":fraction_text(expected),
"coefficient_convention":"ordinary_multi_affine_weight_mu_C_manuscript_equation_9",
"generation_check":"exact_formula_parameter_bounds_and_equal_exchange_ratios_passed",
"global_property_status":"Lorentzian_interior_property_proved_in_manuscript",
"proof_assistant_verified":False,"source_tex":SOURCE_TEX,
"source_tex_sha256":source_hash,"dataset_version":VERSION,
"record_provenance":"analytically_constructed_rational_coefficient_not_observation",
})
return rows
def extract_square(n: int) -> tuple[int,int]:
"""Return a,b with sqrt(n)=a*sqrt(b), b squarefree."""
a=1;b=n;p=2
while p*p<=b:
while b%(p*p)==0:
a*=p;b//=p*p
p+=1
return a,b
def profile_records(source_hash: str) -> list[dict]:
rows=[]
for d in range(1,129):
for label,q in [("1/2",Fraction(1,2)),("1",Fraction(1)),("2",Fraction(2)),("infinity",None)]:
rational=None
if label=="1/2":rational=Fraction(2**(d-1),d)
elif label=="1":rational=Fraction(2**(d-1))
elif label=="infinity":rational=Fraction(d*2**(d-1))
else:
a,b=extract_square(d)
if b==1:rational=Fraction(2**(d-1)*a)
if rational is not None:
expression=fraction_text(rational)
else:
a,b=extract_square(d)
expression=f"{2**(d-1)*a}*sqrt({b})"
rows.append({
"record_id":f"d{d}_q{label.replace('/','_')}","exchange_distance":d,
"q_label":label,"q_is_infinite":q is None,
"q_numerator":None if q is None else str(q.numerator),
"q_denominator":None if q is None else str(q.denominator),
"profile_kind":"quasi_norm" if label=="1/2" else "norm",
"constant_is_rational":rational is not None,
"constant_rational_numerator":None if rational is None else str(rational.numerator),
"constant_rational_denominator":None if rational is None else str(rational.denominator),
"constant_expression":expression,
"general_formula":"2^(d-1)*d^(1-1/q); 1/infinity=0",
"inequality":"mu(S)*mu(T) <= K_(d,q)*unnormalized_lq_exchange_profile",
"sharpness_status":"optimal_supremum_proved_in_manuscript",
"proof_status":STATUS,"historical_priority":PRIORITY,
"proof_assistant_verified":False,"source_theorem":"Theorem 3",
"source_pdf":SOURCE_PDF,"source_tex":SOURCE_TEX,
"source_tex_sha256":source_hash,"dataset_version":VERSION,
"record_provenance":"exact_analytic_constant_not_empirical_measurement",
})
return rows
def infer_field_schema(rows: list[dict]) -> dict:
out={}
keys=list(rows[0])
assert all(list(r)==keys for r in rows)
for key in keys:
values=[r[key] for r in rows]
nonnull=[v for v in values if v is not None]
sample=nonnull[0] if nonnull else None
if isinstance(sample,bool):typ="boolean"
elif isinstance(sample,int):typ="integer"
elif isinstance(sample,str):typ="string"
elif isinstance(sample,list):typ="array<integer>" if sample and isinstance(sample[0],int) else "array<string>"
else:raise TypeError((key,type(sample)))
out[key]={"type":typ,"nullable":any(v is None for v in values)}
return out
def main(repository: Path, manuscript: Path):
data=repository/"data"
source=manuscript.read_bytes();source_hash=hashlib.sha256(source).hexdigest()
text=source.decode("utf-8")
for anchor in ["main","constant","lq","telescoping","rational","refutation","ternarycriterion","transferregion","extension"]:
assert "\\label{"+anchor+"}" in text
configs={"claims":claim_records(source_hash),"family_coefficients":family_records(source_hash),"exchange_profiles":profile_records(source_hash)}
outputs={}
for name,rows in configs.items():
assert len({r['record_id'] for r in rows})==len(rows)
target=data/name/"train.jsonl";target.parent.mkdir(parents=True,exist_ok=True)
target.write_text("".join(json.dumps(r,ensure_ascii=False,separators=(',',':'))+"\n" for r in rows),encoding="utf-8")
for line in target.read_text().splitlines():json.loads(line)
outputs[name]={"path":f"data/{name}/train.jsonl","split":"train","rows":len(rows),"sha256":hashlib.sha256(target.read_bytes()).hexdigest(),"features":infer_field_schema(rows)}
schema={
"dataset_version":VERSION,
"collection_kind":"machine_readable_mathematical_research_records",
"split_semantics":"train is a Hub collection partition only; no training/evaluation split or model benchmark is claimed",
"source_tex":SOURCE_TEX,"source_pdf":SOURCE_PDF,"source_tex_sha256":source_hash,
"configs":outputs,
"exact_number_encoding":{
"rational":"A reduced rational uses base-10 string numerator and positive base-10 string denominator; exact strings are numerator/denominator, including /1.",
"large_integers":"Numerators and denominators are strings to avoid 64-bit overflow and float rounding.",
"irrational_profile_constant":"For q=2 and nonsquare d, constant_expression is an exact integer times sqrt(squarefree_integer); rational numerator and denominator are null.",
"infinite_q":"q_is_infinite is true and q_numerator/q_denominator are null; use q_label=infinity.",
},
"family_formula":"Let j=|C intersect A|, t=s*lambda, h_j=t^(j*(j-1)/2), g_j=s^j*h_j. mu(C)=h_j/binom(r,j) if u notin C, otherwise (j+1)*g_j/binom(r,j).",
"family_parameters":["polynomial_exchange_counterexample: s=1/4, lambda=3/2","strict_reference: s=1/10, lambda=9/5","near_optimal_N_100r: N=100*r, s=1/(2*N), lambda=2-1/N"],
"labels":"Family records are one-based: A=1..r, B=r+1..2r-1, u=2r.",
"status":"The all-rank theorem is established by the included manuscript proof; generation checks verify formulas and finite identities only. No proof-assistant formalization or exhaustive novelty claim.",
"generation_script":"scripts/generate_dataset.py",
}
(data/"dataset_schema.json").write_text(json.dumps(schema,indent=2,ensure_ascii=False)+"\n",encoding="utf-8")
print(json.dumps({"configs":{k:{"rows":v['rows'],"path":v['path']} for k,v in outputs.items()},"schema":"data/dataset_schema.json","source_tex_sha256":source_hash},indent=2))
if __name__=="__main__":
parser=argparse.ArgumentParser()
parser.add_argument("--repository",type=Path,default=Path(__file__).resolve().parents[1])
parser.add_argument("--manuscript-source",type=Path,default=None)
args=parser.parse_args()
manuscript=args.manuscript_source or args.repository/SOURCE_TEX
main(args.repository,manuscript)
|