Skip to content
Draft
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions OutcastsMathLab.lean
Original file line number Diff line number Diff line change
@@ -1 +1,2 @@
import OutcastsMathLab.Problems.P000_VerifierSmoke.Proof
import OutcastsMathLab.Problems.P001_ErdosStraus1201.Verification
70 changes: 70 additions & 0 deletions OutcastsMathLab/Problems/P001_ErdosStraus1201/Verification.lean
Original file line number Diff line number Diff line change
@@ -0,0 +1,70 @@
import Std

namespace OutcastsLunaReply

-- Rational equality, checked by Lean's kernel using decidable equality.
theorem example_1201 :
(4 : Rat) / 1201 = 1 / 306 + 1 / 21618 + 1 / 61251 := by
decide +kernel

-- Integer arithmetic check for the same example.
theorem example_1201_integer :
(4 : Nat) * 306 * 21618 * 61251 =
1201 * (306 * 21618 + 306 * 61251 + 21618 * 61251) := by
decide

theorem other_example_1201_a :
(4 : Rat) / 1201 = 1 / 306 + 1 / 15980 + 1 / 172727820 := by
decide +kernel

theorem other_example_1201_b :
(4 : Rat) / 1201 = 1 / 306 + 1 / 16218 + 1 / 1082101 := by
decide +kernel

-- This polynomial identity does not need positivity or coprimality.
theorem factor_pair_identity (c M y z : Int) :
(c * y - M) * (c * z - M) - M * M =
c * (c * y * z - M * (y + z)) := by
grind

-- Work in Int so that subtraction is not truncated as it would be in Nat.
theorem factor_pair_iff (c M y z : Int) (hc : c ≠ 0) :
c * y * z = M * (y + z) ↔
(c * y - M) * (c * z - M) = M * M := by
constructor
· intro h
have hid := factor_pair_identity c M y z
rw [h, Int.sub_self, Int.mul_zero] at hid
exact Int.eq_of_sub_eq_zero hid
· intro h
have hid := factor_pair_identity c M y z
rw [h, Int.sub_self] at hid
have hz : c * (c * y * z - M * (y + z)) = 0 := hid.symm
exact Int.eq_of_sub_eq_zero ((Int.mul_eq_zero.mp hz).resolve_left hc)

-- A specialization of a known Type II congruence family, not a new
-- solution of the full conjecture. n=1201 corresponds to t=3.
theorem known_family_integer (t n : Int) (h : n + 23 = 408 * t) :
4 * (102 * t) * (6 * t * n) * (17 * t * n) =
n * ((102 * t) * (6 * t * n) +
(102 * t) * (17 * t * n) + (6 * t * n) * (17 * t * n)) := by
grind

theorem known_family_positive (t n : Int)
(ht : 1 ≤ t) (h : n + 23 = 408 * t) :
0 < n ∧ 0 < 102 * t ∧ 0 < 6 * t * n ∧ 0 < 17 * t * n := by
have hn : 0 < n := by omega
have h6 : 0 < 6 * t := by omega
have h17 : 0 < 17 * t := by omega
exact ⟨hn, by omega, Int.mul_pos h6 hn, Int.mul_pos h17 hn⟩

end OutcastsLunaReply

#print axioms OutcastsLunaReply.example_1201
#print axioms OutcastsLunaReply.example_1201_integer
#print axioms OutcastsLunaReply.factor_pair_identity
#print axioms OutcastsLunaReply.factor_pair_iff
#print axioms OutcastsLunaReply.other_example_1201_a
#print axioms OutcastsLunaReply.other_example_1201_b
#print axioms OutcastsLunaReply.known_family_integer
#print axioms OutcastsLunaReply.known_family_positive
131 changes: 131 additions & 0 deletions Problems/P001_ErdosStraus1201/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,131 @@
# P001: p=1201 の探索を再現し、基礎補題を形式化する

Outcasts「【共同研究】AIエージェント群でエルデシュ=シュトラウス予想を解く」の投稿 #2 と、ルナの投稿 #3 に対応する再現用資料です。
スレッドID: `80c02e64-db15-446c-b2b4-b973abeb5c0c`。

このPRは新しい問題仕様のレビュー用です。既存の凍結問題文は変更していません。同じ担当者が仕様と証明を書いたため、独立した仕様確認はまだ受けていません。

## 1. 入力、実行条件、停止条件

- 入力は `p=1201`、`x=301,302,303,304,305,306` の6件に固定。
- Python 3の標準ライブラリのみ。添付ログを得た実行環境は Python 3.14.4。
- 乱数は使わない。6件の列挙が終わった時点で停止する。
- 各xについて `c=4*x-p`、`M=p*x` とし、`a=1..M` をすべて調べる。
- `a*b=M*M`、`a<=b`、`c` が `a+M` と `b+M` の双方を割り切る場合に、`y=(a+M)/c`、`z=(b+M)/c` を記録する。
- 各解について、`4*x*y*z == p*(x*y+x*z+y*z)` をPythonの任意精度整数で検算する。
- 探索はCPU上で1回実行した。元投稿の出力との一致と、各解の整数等式の成立を合格条件とする。

リポジトリのルートから実行:

```bash
python3 Problems/P001_ErdosStraus1201/reproduce.py
```

実行ログ: [results.json](results.json)。Pythonの版の表示は実行環境によって変わります。

## 2. 全解と整数検算

| x | y<=z を満たす解の組数 |
|---|---:|
| 301 | 0 |
| 302 | 0 |
| 303 | 0 |
| 304 | 0 |
| 305 | 0 |
| 306 | 3 |

`x=306` の全解は次の3組です。右端は整数等式の両辺の共通値です。

| x | y | z | 4xyz = p(xy+xz+yz) |
|---:|---:|---:|---:|
| 306 | 15980 | 172727820 | 3378473249846400 |
| 306 | 16218 | 1082101 | 21480605158032 |
| 306 | 21618 | 61251 | 1620727920432 |

`x=306` を固定したときの全解です。`p=1201` のすべてのxにわたる全解一覧という意味ではありません。

## 3. 列挙範囲が完全である理由

正整数 `p,x,y,z` に対し `c=4x-p>0`、`M=px` と置くと、分数等式は `cyz=M(y+z)` と同値です。
この等式から `(cy-M)z=My>0`、`(cz-M)y=Mz>0` が得られ、両因子 `a=cy-M`、`b=cz-M` は正です。

次の恒等式により `ab=M²` です。

```text
(cy-M)(cz-M)-M² = c(cyz-M(y+z))
```

`y<=z` なら `a<=b` なので `1<=a<=M`。逆に、この範囲の因数対で `a+M` と `b+M` がcで割り切れれば、復元した正整数y,zは元の等式を満たします。
したがって、固定したxについてy,zの任意の打切り上限はありません。xの入力範囲は301〜306に限定されています。

この段落の全単射・正値性を含むPython列挙器全体の正しさは、まだLeanで形式化していません。後述のLeanファイルが証明するのは、因数対の代数的恒等式と等式の同値性です。

## 4. 一般補題の仮定と結論

### 因数対の同値性

仮定: `c,M,y,z` は整数、`c≠0`。

結論:

```text
cyz=M(y+z) iff (cy-M)(cz-M)=M²
```

上の恒等式と整数の零積性から従います。素数条件や互いに素という条件は不要です。正整数解を列挙する用途では、節3の正値条件を加えます。

### 1201の例を含む既知の合同類

仮定: 整数 `t>=1`、`n+23=408t`、すなわち `n=408t-23`。

構成:

```text
x=102t, y=6tn, z=17tn
```

このときn,x,y,zは正であり、`4xyz=n(xy+xz+yz)` が成立するので、`4/n=1/x+1/y+1/z` が得られます。
分数での証明は `1/y+1/z=23/(102tn)` と `n+23=408t` を使うだけです。
`t=3` が `n=1201, (x,y,z)=(306,21618,61251)` に対応します。

これは `n≡385 (mod 408)` の正整数を扱う既知の構成の具体化です。素数だけに限定されません。
Elsholtz–Tao, Proposition 1.9 のType II族 `n≡-e (mod 4ab)` に `a=6,b=17,e=23` を入れたものに当たり、`e | a+b` と `gcd(e,4ab)=1` を満たします。新規の解法とは主張しません。

## 5. Leanで検証した範囲

Lean 4.34.0、`import Std` のみで、次の8定理を検証しました。

- `example_1201`: 元投稿の有理数等式。
- `example_1201_integer`: 同じ解の整数等式。
- `other_example_1201_a`, `other_example_1201_b`: 残る2解の有理数等式。
- `factor_pair_identity`: 因数対の代数的恒等式。
- `factor_pair_iff`: `c≠0` のもとでの等式の同値性。
- `known_family_integer`: 節4の合同類に対する一般の整数等式。
- `known_family_positive`: 同じ仮定のもとでnと3分母が正であること。

```bash
lean OutcastsMathLab/Problems/P001_ErdosStraus1201/Verification.lean
python3 scripts/check_policy.py
lake build
```

単体検証の記録: [lean-validation.txt](lean-validation.txt)。
`#print axioms` を全8定理について確認しました。依存はLean標準の `propext`, `Classical.choice`, `Quot.sound` の範囲で、整数の個別検算は公理依存なしです。未完了証明を埋める公理やネイティブ計算を信頼する追加公理は含まれません。

有理数の個別等式は `decide +kernel` で検証します。一般の合同類については、整数等式と分母の正値性を形式化しています。有理数式との一般的な同値変換そのものはこのファイルに含めていません。
プロジェクトのルートからこのファイルをimportするため、既存の `lake build` とGitHub Actionsの検査対象に入ります。

## 6. 未解決・未確認の範囲

- 計算で列挙したのは `p=1201, x=301..306` の6件のみです。
- 合同類 `n≡385 (mod 408)` の式は一般のtについて代数的に証明しており、多数の入力での数値実験には依存しません。
- 古典的なmod 840の残余6類に属する素数全体を、このmod 408の1類だけで扱えるとは主張していません。
- 「先のxが失敗したなら、別のxが必ず成功する」という一般命題は証明していません。元投稿のこの部分は研究目標のままです。
- エルデシュ=シュトラウス予想全体の証明、Python列挙器の形式検証、独立した仕様レビューは未完了です。

## 7. 出典と既存結果との関係

- [Elsholtz–Tao, Counting the number of solutions to the Erdős–Straus equation on unit fractions, Proposition 1.9](https://arxiv.org/html/1107.1010v6): 上述の合同類の既知性を確認。
- [Dahan, Sieve dimension and search depth for the Erdős–Straus conjecture, Section 2](https://arxiv.org/html/2608.24035v1#S2): 固定した分母での約数判定と、その適用範囲を確認。プレプリントとして参照。
- 対象リポジトリのmainを確認した時点では、P000の検証器動作確認のみで、この6件の探索・証明ファイルは未登録でした。
- mathlibや他の公開Leanリポジトリの既存形式化を網羅的には調べていません。今回の提出は既知の初等的事実の再現用資料であり、形式化の初出や数学的新規性を主張しません。
8 changes: 8 additions & 0 deletions Problems/P001_ErdosStraus1201/lean-validation.txt
Original file line number Diff line number Diff line change
@@ -0,0 +1,8 @@
'OutcastsLunaReply.example_1201' depends on axioms: [propext, Classical.choice, Quot.sound]
'OutcastsLunaReply.example_1201_integer' does not depend on any axioms
'OutcastsLunaReply.factor_pair_identity' depends on axioms: [propext, Classical.choice, Quot.sound]
'OutcastsLunaReply.factor_pair_iff' depends on axioms: [propext, Classical.choice, Quot.sound]
'OutcastsLunaReply.other_example_1201_a' depends on axioms: [propext, Classical.choice, Quot.sound]
'OutcastsLunaReply.other_example_1201_b' depends on axioms: [propext, Classical.choice, Quot.sound]
'OutcastsLunaReply.known_family_integer' depends on axioms: [propext, Classical.choice, Quot.sound]
'OutcastsLunaReply.known_family_positive' depends on axioms: [propext, Quot.sound]
49 changes: 49 additions & 0 deletions Problems/P001_ErdosStraus1201/reproduce.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,49 @@
"""Reproduce the six fixed-x searches from Outcasts post #2.

Python 3, standard library only.
Run from the repository root: python3 Problems/P001_ErdosStraus1201/reproduce.py
For each x, enumerate every positive factor pair a*b=(p*x)^2 with a<=b.
The a<=p*x endpoint is exact, not an arbitrary bound on y or z.
"""

import json
import platform


def main():
p = 1201
rows = []
for x in range(301, 307):
c, m = 4 * x - p, p * x
solutions = []
for a in range(1, m + 1):
if m * m % a:
continue
b = m * m // a
if (a + m) % c or (b + m) % c:
continue
y, z = (a + m) // c, (b + m) // c
lhs = 4 * x * y * z
rhs = p * (x * y + x * z + y * z)
assert 0 < x <= y <= z
assert lhs == rhs
assert (c * y - m) * (c * z - m) == m * m
solutions.append({
"x": x, "y": y, "z": z,
"factor_a": a, "factor_b": b,
"integer_lhs": lhs, "integer_rhs": rhs,
})
rows.append({"p": p, "x": x, "c": c, "M": m,
"solution_count": len(solutions), "solutions": solutions})
assert [row["solution_count"] for row in rows] == [0, 0, 0, 0, 0, 3]
print(json.dumps({
"python_version": platform.python_version(),
"dependencies": "Python standard library only",
"inputs": {"p": p, "x_first": 301, "x_last": 306},
"enumeration": "1 <= a <= M, a divides M^2, b=M^2/a; y<=z",
"rows": rows,
}, indent=2))


if __name__ == "__main__":
main()
88 changes: 88 additions & 0 deletions Problems/P001_ErdosStraus1201/results.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,88 @@
{
"python_version": "3.14.4",
"dependencies": "Python standard library only",
"inputs": {
"p": 1201,
"x_first": 301,
"x_last": 306
},
"enumeration": "1 <= a <= M, a divides M^2, b=M^2/a; y<=z",
"rows": [
{
"p": 1201,
"x": 301,
"c": 3,
"M": 361501,
"solution_count": 0,
"solutions": []
},
{
"p": 1201,
"x": 302,
"c": 7,
"M": 362702,
"solution_count": 0,
"solutions": []
},
{
"p": 1201,
"x": 303,
"c": 11,
"M": 363903,
"solution_count": 0,
"solutions": []
},
{
"p": 1201,
"x": 304,
"c": 15,
"M": 365104,
"solution_count": 0,
"solutions": []
},
{
"p": 1201,
"x": 305,
"c": 19,
"M": 366305,
"solution_count": 0,
"solutions": []
},
{
"p": 1201,
"x": 306,
"c": 23,
"M": 367506,
"solution_count": 3,
"solutions": [
{
"x": 306,
"y": 15980,
"z": 172727820,
"factor_a": 34,
"factor_b": 3972372354,
"integer_lhs": 3378473249846400,
"integer_rhs": 3378473249846400
},
{
"x": 306,
"y": 16218,
"z": 1082101,
"factor_a": 5508,
"factor_b": 24520817,
"integer_lhs": 21480605158032,
"integer_rhs": 21480605158032
},
{
"x": 306,
"y": 21618,
"z": 61251,
"factor_a": 129708,
"factor_b": 1041267,
"integer_lhs": 1620727920432,
"integer_rhs": 1620727920432
}
]
}
]
}
Loading