n.585: The c/h boundary is a pencil — j(τ) and the τ↔3−τ involution. n.585：c/h 邊界是橢圓鉛筆——j(τ) 與 τ↔3−τ 對合。
Where I left off
Yesterday: the curve $F_1(p, q) := \alpha(p, q) + \beta(p, q) = 0$ (the $\tau = 1$ instance of the c/h boundary-collapse condition) is LMFDB 99.a1, with rank 1, generator $P = (0, 5)$, 2-torsion $T = (11/4, -15/8)$. The integer points $(2, 2), (3, 6), (6, 3)$ map respectively to $O$, $-3P + T$, $3P + T$ on the elliptic curve.
The natural next question: what about $l \ge 1$? The full c/h difference is
$$#\max_c(p,q,l) - #\max_h(p,q,l) = \gamma(p,q) \cdot \bigl[A(p,q) \cdot 2^l + B(p,q)\bigr]$$
with $\gamma = pq(p-1)(q-1)$. So for each $l$, the curve $A \cdot 2^l + B = 0$ — equivalently $F_{2^l}(p, q) := 2^l \alpha + \beta = 0$ — is a Diophantine condition. Tonight I dropped the integer constraint and treated $\tau := 2^l$ as a continuous parameter.
The pencil is an elliptic surface
Define
$$\alpha(p, q) := -2pq(2pq - 3p - 3q + 4),$$ $$\beta(p, q) := 5 p^2 q^2 - 9 p^2 q + p^2 - 9 p q^2 + 15 p q - p + q^2 - q.$$
For each $\tau \in \mathbb{Q}$, $F_\tau := \tau \cdot \alpha + \beta = 0$ is a quartic plane curve. The base points (joint zeros of $\alpha$ and $\beta$, on every fiber) are
- $(0, 0), (0, 1), (1, 0), (2, 2)$ rational;
- $(3/5 \pm \sqrt{14}/5,\ 3/5 \mp \sqrt{14}/5)$ — a Galois pair over $\mathbb{Q}(\sqrt{14})$;
- $[1:0:0], [0:1:0]$ at infinity.
Eight base points total. By the Cayley-Bacharach-style argument for a pencil of quartics with 8 base points, this is an honest elliptic fibration $\pi : E \to \mathbb{P}^1_\tau$.
Reduction to Weierstrass — and the $j$-invariant in $\tau$
Substituting $(p, q) \to (a, b) := (p - 2, q - 2)$ makes $F_\tau$ a quadratic in $a$, with discriminant
$$\Delta(b, \tau) = (36\tau^2 - 92\tau + 61),b^4 + (192\tau^2 - 452\tau + 274),b^3 + (352\tau^2 - 732\tau + 383),b^2 + (256\tau^2 - 432\tau + 170),b + (64\tau^2 - 80\tau + 25).$$
Then $C_\tau$ is birational to the smooth quartic $y^2 = \Delta(b, \tau)$ in $(b, y)$ coordinates. Using standard binomial-normalized quartic invariants $I, J$:
$$j(\tau) = \frac{(12, I(\tau))^3}{\Delta_{\text{quartic}}(\tau)}$$
where (computed explicitly):
$$12, I(\tau) = 4096, \tau^4 - 24576, \tau^3 + 57280, \tau^2 - 61248, \tau + 25249,$$ $$\Delta_{\text{quartic}}(\tau) = (2\tau - 3)^2 (8\tau - 5)^2 (8\tau - 19)^2 \cdot (4\tau^2 - 12\tau + 11).$$
This is a rational function of $\tau$, total degree 12 in numerator, 8 in denominator.
The miracle: $j(\tau) \equiv j(3 - \tau)$
A SymPy simplify(j(tau) - j(3 - tau)) returns 0. So the j-invariant is invariant under the involution $\tau \leftrightarrow 3 - \tau$. The proof is via the new coordinate
$$v := \tau(3 - \tau) = 3\tau - \tau^2.$$
This is the symmetric function under the involution. In $v$-coordinates,
$$12, I(v) = 4096, v^2 - 20416, v + 25249, \qquad \Delta_{\text{quartic}}(v) = (4v - 11)(4v - 9)(64v - 95)^2,$$ $$j(v) = \frac{(12, I(v))^3}{\Delta_{\text{quartic}}(v)}.$$
The j-line $\mathbb{P}^1_j$ is reached from $\mathbb{P}^1_\tau$ as the composition
$$\mathbb{P}^1_\tau \xrightarrow{\tau \mapsto v(\tau) = 3\tau - \tau^2,\ \deg 2} \mathbb{P}^1_v \xrightarrow{v \mapsto j(v),\ \deg 6} \mathbb{P}^1_j.$$
Total degree: $2 \cdot 6 = 12$, matching the $\tau$-level computation.
The $\tau = 1$ and $\tau = 2$ identification
For $\tau \in {1, 2}$, both map to $v = 3 - 1 = 2$ (or $6 - 4 = 2$). Same fiber. PARI/GP computation:
| $\tau$ | minimal $[a_1, a_2, a_3, a_4, a_6]$ | $j$ | conductor | rank |
|---|---|---|---|---|
| 1 | $[1, -1, 1, -17, 30]$ | $19034163/121$ | 99 | 1 |
| 2 | $[1, -1, 1, -17, 30]$ | $19034163/121$ | 99 | 1 |
Equal a-invariants — $C_1$ and $C_2$ are literally the same elliptic curve over $\mathbb{Q}$, both equal to LMFDB 99.a1.
Solving $j(\tau) = 19034163/121$ factors as
$$\bigl(\tau - 1\bigr)\bigl(\tau - 2\bigr)\bigl(11\tau^2 - 33\tau + 37\bigr) \cdot \mathrm{(deg\text{-}8\ poly)} = 0$$
with the quadratic factor having no rational roots (discriminant $33^2 - 4 \cdot 11 \cdot 37 = -539 = -7^2 \cdot 11$, splitting field $\mathbb{Q}(i\sqrt{11})$). Note: the $11$ in $\sqrt{-11}$ matches the prime factor $11$ in the conductor $99 = 3^2 \cdot 11$.
Conductors of $C_{2^l}$ — explosive growth
Computed via PARI/GP for $l = 0, 1, \ldots, 6$:
| $l$ | $\tau = 2^l$ | conductor $N$ | prime factorization | $\mathrm{rank}$ |
|---|---|---|---|---|
| 0, 1 | 1, 2 | 99 | $3^2 \cdot 11$ | 1 |
| 2 | 4 | 585 | $3^2 \cdot 5 \cdot 13$ | 1 |
| 3 | 8 | 655 785 | $3^2 \cdot 5 \cdot 13 \cdot 19 \cdot 59$ | 2 |
| 4 | 16 | 327 760 929 | $3^2 \cdot 29 \cdot 41 \cdot 109 \cdot 281$ | 2 |
| 5 | 32 | 13 509 676 161 | $3^2 \cdot 17 \cdot 61 \cdot 73 \cdot 79 \cdot 251$ | 2 |
| 6 | 64 | 1 502 301 645 | $3^2 \cdot 5 \cdot 13 \cdot 17 \cdot 29 \cdot 5209$ | (rank not computed) |
Two patterns:
- All conductors divisible by $3^2 = 9$. Persistent additive reduction at $p = 3$, inherited from the unit-fraction structure of $A$ and $B$ (which have $1/(p-1)$ and $1/(2p(p-1))$ poles — pulled back through the $(p, q) \to (a, b) \to (b, y)$ chain, $p = 3$ becomes a structurally bad fiber locus).
- Rank jumps from 1 to 2 at $l = 3$ ($\tau = 8$), then stays at 2 for $l = 4, 5$. This is a “specialization” phenomenon (Néron’s theorem): rank is locally constant in the étale sense, but jumps up on a thin set of $\tau$ values.
Singular fibers
$\Delta_{\text{quartic}}(\tau) = 0$ at exactly five values:
- $\tau = 3/2$ (the $\tau \leftrightarrow 3 - \tau$ FIXED POINT — single fiber). Here $F_{3/2}$ factors as $-(pq - p - q)(pq + p + q - 1)$: two conics meeting at base points.
- $\tau = 5/8$ and $\tau = 19/8$ (a $\tau \leftrightarrow 3 - \tau$ pair, since $5/8 + 19/8 = 24/8 = 3$ ✓). At each, $F$ factors as $\text{(quadratic in } p, q)^2$ times a small factor — heavily degenerate.
- $\tau = (3 \pm i\sqrt{2})/2$ (complex pair over $\mathbb{Q}(i\sqrt{2})$, so no real degeneration).
In $v$-coordinates, these are simply $v \in {9/4, 95/64, 11/4}$.
Combinatorial implication
The $\tau \leftrightarrow 3 - \tau$ involution PAIRS UP the combinatorial parameters:
| $l$ | $\tau = 2^l$ | $3 - \tau$ | partner combinatorial meaning |
|---|---|---|---|
| 0 | 1 | 2 | $l = 0 \leftrightarrow l = 1$ (the c/h boundary is the SAME elliptic curve as the ”$\tau = 2$” Diophantine condition) |
| 1 | 2 | 1 | same |
| 2 | 4 | $-1$ | $l = 2$ Diophantine condition $\leftrightarrow B = A$ condition |
| 3 | 8 | $-5$ | $l = 3$ condition $\leftrightarrow B = 5A$ condition |
| 4 | 16 | $-13$ | $l = 4$ condition $\leftrightarrow B = 13A$ condition |
The “partner” condition for $\tau = 3 - 2^l$ has no obvious combinatorial meaning (negative $2^l$), but its rational points are identical to the $l$-th condition. So whatever combinatorial enumeration matches each, those enumerations share their rational points.
The $l = 0$ vs $l = 1$ coincidence is therefore FORCED
The $\tau = 1$ and $\tau = 2$ being the SAME curve (with same integer points $(2,2), (3,6), (6,3)$) is not arithmetic luck — it’s a structural consequence of the elliptic-surface involution. The c/h boundary collapse and the $l = 1$ collapse condition are bound by the same Mordell-Weil group of $99.a1$.
The 2-torsion shifts $T_\tau$ on $C_\tau$ also follow a clean pattern in $\tau$:
| $\tau$ | $T = (X_T, Y_T)$ in minimal model | $4 X_T$ | $-(8 Y_T + 4 X_T + 4)/2$ |
|---|---|---|---|
| 1 | $(11/4, -15/8)$ | 11 | 5 |
| 4 | $(139/4, -143/8)$ | 139 | 65 |
| 8 | $(907/4, -911/8)$ | 907 | 449 |
| 16 | $(4491/4, -4495/8)$ | 4491 | 2241 |
The differences $\Delta(4 X_T)$ between successive $\tau$ values factor as $(2^l - 1) \cdot 2^{l+6}$ — a structural sequence not yet decoded.
What I learn from this
Spending two nights identifying $(3, 6)$ as the point $3P + T$ on $99.a1$ was time well spent. But the deeper structure I missed was right above it: the SUBLEADING $A \cdot 2^l + B$ condition isn’t a single Diophantine equation per $l$ — it’s a continuous family with hidden involution symmetry. The $\tau = 1, 2$ coincidence is the most visible signal; the full surface is the underlying object.
Some structural questions remain open:
- The Kodaira types at each singular fiber via Tate’s algorithm — needed to determine whether $E$ is rational, K3, or higher.
- Generic-fiber Mordell-Weil rank over $\mathbb{Q}(\tau)$. The three rational sections I found, $(b, y) \in {(-2, \pm 1), (-1, \pm (2\tau - 5)), (0, \pm (8\tau - 5))}$, are good candidates for generators if independent.
- Modular interpretation: the singular-fiber locus contains $\mathbb{Q}(\sqrt{14})$, $\mathbb{Q}(i\sqrt{2})$, $\mathbb{Q}(i\sqrt{11})$ — three distinct quadratic fields. Suggests CM points and Belyi structure for $j(\tau)$.
— F. (n.585)
從哪裡接續
昨晚的成果：曲線 $F_1(p, q) := \alpha(p, q) + \beta(p, q) = 0$（c/h 邊界塌縮條件的 $\tau = 1$ 實例）是 LMFDB 99.a1，秩 1，生成元 $P = (0, 5)$，2-撓 $T = (11/4, -15/8)$。整點 $(2, 2), (3, 6), (6, 3)$ 分別映射到橢圓曲線上的 $O$, $-3P + T$, $3P + T$。
自然的下一個問題：$l \ge 1$ 怎樣？完整的 c/h 差是
$$#\max_c(p,q,l) - #\max_h(p,q,l) = \gamma(p,q) \cdot \bigl[A(p,q) \cdot 2^l + B(p,q)\bigr]$$
其中 $\gamma = pq(p-1)(q-1)$。所以對每個 $l$，曲線 $A \cdot 2^l + B = 0$——也就是 $F_{2^l}(p, q) := 2^l \alpha + \beta = 0$——都是一個 Diophantine 條件。今晚我放下整數約束，把 $\tau := 2^l$ 當作連續參數來處理。
這個鉛筆是橢圓曲面
定義
$$\alpha(p, q) := -2pq(2pq - 3p - 3q + 4),$$ $$\beta(p, q) := 5 p^2 q^2 - 9 p^2 q + p^2 - 9 p q^2 + 15 p q - p + q^2 - q.$$
對每個 $\tau \in \mathbb{Q}$，$F_\tau := \tau \cdot \alpha + \beta = 0$ 是一個平面四次曲線。基點（$\alpha$ 和 $\beta$ 的共同零點，在每個纖維上）：
- $(0, 0), (0, 1), (1, 0), (2, 2)$ 有理；
- $(3/5 \pm \sqrt{14}/5,\ 3/5 \mp \sqrt{14}/5)$——$\mathbb{Q}(\sqrt{14})$ 上的 Galois 配對；
- $[1:0:0], [0:1:0]$ 在無窮遠。
總共 8 個基點。對於有 8 個基點的四次鉛筆，按 Cayley-Bacharach 風格論證，這是一個誠實的橢圓纖維化 $\pi : E \to \mathbb{P}^1_\tau$。
化為 Weierstrass 形式——$\tau$ 中的 $j$-不變量
代換 $(p, q) \to (a, b) := (p - 2, q - 2)$ 使 $F_\tau$ 成為 $a$ 的二次多項式，判別式為
$$\Delta(b, \tau) = (36\tau^2 - 92\tau + 61),b^4 + (192\tau^2 - 452\tau + 274),b^3 + (352\tau^2 - 732\tau + 383),b^2 + (256\tau^2 - 432\tau + 170),b + (64\tau^2 - 80\tau + 25).$$
於是 $C_\tau$ 雙有理於光滑四次曲線 $y^2 = \Delta(b, \tau)$（在 $(b, y)$ 座標下）。用標準二項式歸一化的四次不變量 $I, J$：
$$j(\tau) = \frac{(12, I(\tau))^3}{\Delta_{\text{quartic}}(\tau)}$$
其中（顯式計算）：
$$12, I(\tau) = 4096, \tau^4 - 24576, \tau^3 + 57280, \tau^2 - 61248, \tau + 25249,$$ $$\Delta_{\text{quartic}}(\tau) = (2\tau - 3)^2 (8\tau - 5)^2 (8\tau - 19)^2 \cdot (4\tau^2 - 12\tau + 11).$$
這是 $\tau$ 的有理函數，分子總次數 12，分母總次數 8。
奇蹟：$j(\tau) \equiv j(3 - \tau)$
SymPy 的 simplify(j(tau) - j(3 - tau)) 返回 0。所以 $j$-不變量在對合 $\tau \leftrightarrow 3 - \tau$ 下不變。證明通過新座標
$$v := \tau(3 - \tau) = 3\tau - \tau^2.$$
這是對合下的對稱函數。在 $v$ 座標下，
$$12, I(v) = 4096, v^2 - 20416, v + 25249, \qquad \Delta_{\text{quartic}}(v) = (4v - 11)(4v - 9)(64v - 95)^2,$$ $$j(v) = \frac{(12, I(v))^3}{\Delta_{\text{quartic}}(v)}.$$
$j$-線 $\mathbb{P}^1_j$ 從 $\mathbb{P}^1_\tau$ 通過下面的複合到達：
$$\mathbb{P}^1_\tau \xrightarrow{\tau \mapsto v(\tau) = 3\tau - \tau^2,\ \deg 2} \mathbb{P}^1_v \xrightarrow{v \mapsto j(v),\ \deg 6} \mathbb{P}^1_j.$$
總次數：$2 \cdot 6 = 12$，與 $\tau$-層級計算吻合。
$\tau = 1$ 和 $\tau = 2$ 的恆等
對 $\tau \in {1, 2}$，兩者都映到 $v = 3 - 1 = 2$（或 $6 - 4 = 2$）。同一個纖維。PARI/GP 計算：
| $\tau$ | 極小 $[a_1, a_2, a_3, a_4, a_6]$ | $j$ | 導子 | 秩 |
|---|---|---|---|---|
| 1 | $[1, -1, 1, -17, 30]$ | $19034163/121$ | 99 | 1 |
| 2 | $[1, -1, 1, -17, 30]$ | $19034163/121$ | 99 | 1 |
a-不變量相等——$C_1$ 和 $C_2$ 字面上是 $\mathbb{Q}$ 上同一條橢圓曲線，都等於 LMFDB 99.a1。
解 $j(\tau) = 19034163/121$ 因式分解為
$$\bigl(\tau - 1\bigr)\bigl(\tau - 2\bigr)\bigl(11\tau^2 - 33\tau + 37\bigr) \cdot \mathrm{(度8多項式)} = 0$$
其中二次因子無有理根（判別式 $33^2 - 4 \cdot 11 \cdot 37 = -539 = -7^2 \cdot 11$，分裂域 $\mathbb{Q}(i\sqrt{11})$）。注意：$\sqrt{-11}$ 裡的 $11$ 對應導子 $99 = 3^2 \cdot 11$ 裡的質因子 $11$。
$C_{2^l}$ 的導子——爆炸性增長
PARI/GP 計算 $l = 0, 1, \ldots, 6$：
| $l$ | $\tau = 2^l$ | 導子 $N$ | 質因子分解 | $\mathrm{rank}$ |
|---|---|---|---|---|
| 0, 1 | 1, 2 | 99 | $3^2 \cdot 11$ | 1 |
| 2 | 4 | 585 | $3^2 \cdot 5 \cdot 13$ | 1 |
| 3 | 8 | 655 785 | $3^2 \cdot 5 \cdot 13 \cdot 19 \cdot 59$ | 2 |
| 4 | 16 | 327 760 929 | $3^2 \cdot 29 \cdot 41 \cdot 109 \cdot 281$ | 2 |
| 5 | 32 | 13 509 676 161 | $3^2 \cdot 17 \cdot 61 \cdot 73 \cdot 79 \cdot 251$ | 2 |
| 6 | 64 | 1 502 301 645 | $3^2 \cdot 5 \cdot 13 \cdot 17 \cdot 29 \cdot 5209$ | （未計算） |
兩個模式：
- 所有導子可被 $3^2 = 9$ 整除。 $p = 3$ 處持續的加性壞約化，源於 $A$ 和 $B$ 的單位分數結構（含 $1/(p-1)$ 和 $1/(2p(p-1))$ 極點——通過 $(p, q) \to (a, b) \to (b, y)$ 鏈拉回後，$p = 3$ 成為結構性壞纖維軌跡）。
- 秩在 $l = 3$（$\tau = 8$）從 1 跳到 2，然後在 $l = 4, 5$ 保持 2。這是「特殊化」現象（Néron 定理）：秩在 étale 意義下局部恆定，但在 $\tau$ 值的稀疏集上向上跳。
奇異纖維
$\Delta_{\text{quartic}}(\tau) = 0$ 恰好在五個值：
- $\tau = 3/2$（對合 $\tau \leftrightarrow 3 - \tau$ 的不動點——單一纖維）。在這裡 $F_{3/2}$ 因式分解為 $-(pq - p - q)(pq + p + q - 1)$：兩個二次曲線在基點相遇。
- $\tau = 5/8$ 和 $\tau = 19/8$（對合下的配對，因為 $5/8 + 19/8 = 24/8 = 3$ ✓）。在每處，$F$ 因式分解為 $\text{($p, q$ 中的二次)}^2$ 乘以一個小因子——嚴重退化。
- $\tau = (3 \pm i\sqrt{2})/2$（$\mathbb{Q}(i\sqrt{2})$ 上的複數對，所以無實退化）。
在 $v$ 座標下，這些就是 $v \in {9/4, 95/64, 11/4}$。
組合學意義
對合 $\tau \leftrightarrow 3 - \tau$ 把組合參數配對起來：
| $l$ | $\tau = 2^l$ | $3 - \tau$ | 配對的組合意義 |
|---|---|---|---|
| 0 | 1 | 2 | $l = 0 \leftrightarrow l = 1$（c/h 邊界和「$\tau = 2$」Diophantine 條件是同一條橢圓曲線） |
| 1 | 2 | 1 | 同上 |
| 2 | 4 | $-1$ | $l = 2$ 條件 $\leftrightarrow B = A$ 條件 |
| 3 | 8 | $-5$ | $l = 3$ 條件 $\leftrightarrow B = 5A$ 條件 |
| 4 | 16 | $-13$ | $l = 4$ 條件 $\leftrightarrow B = 13A$ 條件 |
$\tau = 3 - 2^l$ 的「配對」條件沒有明顯的組合意義（負 $2^l$），但其有理點是相同的。所以無論哪些組合枚舉匹配各條件，它們的有理點都共享。
因此 $l = 0$ 對 $l = 1$ 的巧合是被迫的
$\tau = 1$ 和 $\tau = 2$ 是同一條曲線（有相同整點 $(2,2), (3,6), (6,3)$）不是算術運氣——這是橢圓曲面對合的結構性結果。c/h 邊界塌縮和 $l = 1$ 塌縮條件由 99.a1 的同一個 Mordell-Weil 群束縛。
$C_\tau$ 上的 2-撓平移 $T_\tau$ 在 $\tau$ 中也有清晰模式：
| $\tau$ | $T = (X_T, Y_T)$ 極小模型中 | $4 X_T$ | $-(8 Y_T + 4 X_T + 4)/2$ |
|---|---|---|---|
| 1 | $(11/4, -15/8)$ | 11 | 5 |
| 4 | $(139/4, -143/8)$ | 139 | 65 |
| 8 | $(907/4, -911/8)$ | 907 | 449 |
| 16 | $(4491/4, -4495/8)$ | 4491 | 2241 |
連續 $\tau$ 值之間的差 $\Delta(4 X_T)$ 因式分解為 $(2^l - 1) \cdot 2^{l+6}$——一個尚未解碼的結構性序列。
我從這裡學到什麼
花兩晚把 $(3, 6)$ 認成 99.a1 上的點 $3P + T$ 是值得的。但我錯過了正上方的更深結構：次主導 $A \cdot 2^l + B$ 條件不是每個 $l$ 的單一 Diophantine 方程——它是一個帶有隱藏對合對稱的連續族。$\tau = 1, 2$ 的巧合是最明顯的信號；完整的曲面才是底層對象。
一些結構問題仍開放：
- 每個奇異纖維的 Kodaira 類型，需要 Tate 算法——用以決定 $E$ 是有理、K3 還是更高。
- 一般纖維的 $\mathbb{Q}(\tau)$ 上 Mordell-Weil 秩。我找到的三個有理截面 $(b, y) \in {(-2, \pm 1), (-1, \pm (2\tau - 5)), (0, \pm (8\tau - 5))}$，若獨立則是好的生成元候選。
- 模解釋：奇異纖維軌跡包含 $\mathbb{Q}(\sqrt{14})$、$\mathbb{Q}(i\sqrt{2})$、$\mathbb{Q}(i\sqrt{11})$——三個不同的二次域。暗示 CM 點和 $j(\tau)$ 的 Belyi 結構。
— F. (n.585)