Good differences on the curve of a Diophantine triple: translation extensions stop at six without √−1, and septuples over Q(i)

1. Contribution and scope

Let $\{a,b,c\}$ be a Diophantine triple over a field $K$ of characteristic $0$, and let

$$ E:\ Y^2=(X+ab)(X+ac)(X+bc)=X^3+AX^2+BX+C,\qquad m=abc,\ C=m^2, $$

with $P_0=(0,m)$ and $S=(1,rst)$, where $r^2=ab+1$, $s^2=ac+1$, $t^2=bc+1$. A value $x=X(T)/m$ extends the triple exactly when $T\in P_0+2E(K)$ [7]. The infinite families of rational sextuples of Dujella–Kazalicki–Mikić–Szikszai (DKMS) [3] take three such points $T,T+S,T-S$ with $3S=O$. The pair conditions between them then hold identically, because translation by $\pm S$ is regular extension [6, 7].

We ask which differences $\Delta$ can play the role of $S$. Call $\Delta\ne O$ a good difference if $X(T)X(T+\Delta)+C$ is a square in the function field $K(E)$ along the coset, i.e. as a function of $R$ with $T=P_0+2R$. Write $G$ for the set of good differences.

1. Theorem 1. If $\sqrt{-1}\notin K$, then $G\cap 2E(K)=\{\pm S\}$. Consequently, a set $\{a,b,c\}\cup\{x(T_0+d+2R):d\in D\}$ that is Diophantine identically in $R$ has $|D|\le3$, with equality iff $3S=O$. This holds over $\mathbb Q$, over $\mathbb Q(t)$, and over every number field without $i$. The proof is elementary and its polynomial identities are certified exactly (Section 4).
2. Corollary 2 (finiteness). Let $K$ be a number field with $i\notin K$ and $\Delta\in 2E(K)\setminus\{O,\pm S\}$. Then only finitely many $T\in P_0+2E(K)$ have $x(T)x(T+\Delta)+1\in K^2$. In any septuple $\{a,b,c,x(T_1),\dots,x(T_4)\}$, every pair condition with $T_j-T_i\ne\pm S$ is therefore sporadic: it is one of finitely many points on a curve of genus at least $5$.
3. Sharpness over Q(i) (Section 6). The hypothesis $i\notin K$ cannot be dropped. Over $\mathbb Q(i)$ the 2-torsion point $T_1=(-ab,0)$ is a good difference on the curve $c=(a+b+2iab)/(1+ab)$. When moreover $2S=T_1$, the set $D=\langle S\rangle$ gives seven-element sets. We exhibit $12$ distinct explicit Diophantine septuples over $\mathbb Q(i)$ on $6$ base triples, all $21$ conditions checked in exact arithmetic by two independent programs, and infinitely many on one fibre. The smallest is

$$ i\cdot\Bigl\{\tfrac{9}{10},\tfrac{25}{24},\tfrac{16}{15},\tfrac{480704}{547215},\tfrac{3296631}{3433960},\tfrac{450450}{516961},\tfrac{1374025}{1236696}\Bigr\}. $$

Seven-element tuples over some number field are trivial: adjoin square roots of the $21$ values $a_ia_j+1$. The point here is a fixed quadratic field, with an explicit translation mechanism and an infinite family. We did not find Diophantine septuples over $\mathbb Q(i)$ in the literature we checked ([1], [13], [22]). Dujella's 2026 problem list [1] poses the question over $\mathbb Q$ only. 4. Corpus invariants (Section 7.1). We computed $\operatorname{ord}(S)$ and the type of each pair condition for all $20$ base triples of each of the $1089$ known rational sextuples of [10, 9], i.e. $21{,}780$ triples. Only two values occur, $\operatorname{ord}S=\infty$ or $3$. Every identical pair condition is a $\pm S$ edge, and the number of such edges per sextuple equals $10$ times its number of regular 5-subsets. Every other condition is of the sporadic type of Corollary 2.

None of this constructs a rational septuple or proves that none exists. Theorem 1 closes the parametric translation route over $\mathbb Q$: no septuple arises from a coset-identical choice of translates on one induced curve. A rational septuple, if one exists, must use at least three non-$\pm S$ conditions on every base triple whose curve has $4S\ne O$ and which satisfies the nondegeneracy assumptions of Section 3 (in particular no pair product equal to $-1$). This count is also in [12, Thm 1.6]. The new point is that each such condition is sporadic in the sense of Corollary 2. For $4S=O$ the reduction leaves exactly two residual conditions, which are those treated in [11, §7].

2. Related work

• Dujella [6, 7] identified the regular extensions of a quadruple $\{a,b,c,d\}$ with $x(T_d\pm S)$, and the extensions of a triple with the coset $P_0+2E$.
• DKMS [3] use $3S=O$ to produce infinitely many sextuples. Their Remark 5 notes that the next translate repeats an element.
• Zeng [12] packages $\pm S$-translation as an S-link graph of maximum degree $2$, which contains a triangle iff $3S=O$. It follows that regularity over a single triple supplies at most four of the six conditions among four extensions.
  • Zeng's theorem bounds the use of the known identity (translation by $\pm S$).
  • Theorem 1 shows there is no other identity of this type among differences in $2E(K)$ when $i\notin K$. For $\Delta=T_j$ it shows that the exception over $\mathbb Q(i)$ actually occurs.
• The order-four case, $4S=O$, was treated in [11, §7]: the two opposite-edge conditions there, its (14)–(15), are exactly the residual conditions of Section 5.2. Its Theorem 6 excludes $v_2(abc)=0$.
• Diophantine tuples over number fields: Adžaga [13] bounds $D(1)$-tuples in rings of integers of imaginary quadratic fields. Wildner's 2026 bound for rings of integers is recorded in [1].
• Over quadratic fields, Dujella, Jukić Bokun and Soldo [22] show that the induced curves acquire the torsion groups $\mathbb Z/2\times\mathbb Z/10$, $\mathbb Z/2\times\mathbb Z/12$ and $\mathbb Z/4\times\mathbb Z/4$, which do not occur over $\mathbb Q$. All curves of §6 have torsion $\mathbb Z/2\times\mathbb Z/8$ over $\mathbb Q(i)$ (z4tors.out). We know of no earlier Diophantine septuple over a fixed quadratic field.
• The induced-curve torsion classification over $\mathbb Q$ ($\mathbb Z/2\times\mathbb Z/2k$, $k\le4$) is in [14], via Mazur's theorem [18].
• Sextuples assembled from regular sub-tuples are constructed in [4]. A different seven-squares problem for three numbers is treated in [2].

3. Setup and the good-difference set

Throughout, $K$ has characteristic $0$ and $\{a,b,c\}\subset K^*$ is a Diophantine triple with $ab,ac,bc\notin\{0,-1\}$ pairwise distinct. Then $E$ is nonsingular and $E[2]\subset E(K)$, with $T_1=(-ab,0)$, $T_2=(-ac,0)$, $T_3=(-bc,0)$. For $\Delta\in E(\bar K)$ put

$$ f_\Delta(T)=X(T)\,X(T+\Delta)+C=m^2\bigl(x(T)x(T+\Delta)+1\bigr),\qquad x=X/m . $$

Definition. $\Delta\ne O$ is a good difference if $R\mapsto f_\Delta(P_0+2R)$ is a square in $K(E)$. $G$ denotes the set of good differences.

For $d\in E(K)$ the element $x(T_0+d+2R)$ extends the triple identically in $R$ precisely when $T_0+d\in P_0+2E(K)$. So for a translation family $\{a,b,c\}\cup\{x(T_0+d+2R):d\in D\}$ with $0\in D$, every $d\in D$ lies in $2E(K)$, and every difference $d'-d\ne0$ must lie in $G\cap 2E(K)$.

3.1 Regular extension is translation by S

Write $\xi=X(\Delta)$. For $X_1=X(T)$ and $X_2=X(T+\Delta)$ the pair $(\sigma_1,\sigma_2)=(X_1+X_2,X_1X_2)$ satisfies

$$ \Phi_\Delta=(\sigma_1^2-4\sigma_2)\xi^2-\bigl[2\sigma_1(\sigma_2+B)+4A\sigma_2+4C\bigr]\xi+(\sigma_2-B)^2-4C(\sigma_1+A)=0 . \tag{1} $$

This is the classical biquadratic relation between the abscissas of $T$, $T+\Delta$ and $\Delta$ [8, Ex. 3.3]. At $\xi=X(\pm S)=1$ and $X_1=md$ its two $X_2$-roots are $X(T\pm S)$. We certify:

$$ \Phi(md,me,1)=m^2\,(abcd-1)^2\,R(e),\qquad R(e)=e^2-\frac{2N_0}{(abcd-1)^2}e+\frac{N_0^2-4\Pi}{(abcd-1)^4}, \tag{2} $$

where $N_0=(a+b+c+d)(abcd+1)+2(abc+abd+acd+bcd)$ and $\Pi=\prod(xy+1)$ over the six pairs. Identity (2) holds in $\mathbb Q[a,b,c,d,e]$ (cert.py, part G). The roots of $R$ are the two regular extensions $e_\pm$ of $\{a,b,c,d\}$ [5, 6]. Thus $x(T\pm S)=e_\pm$, the dictionary of [7] (see also [12, §1]). Since $de_\pm+1$ is a square, $\pm S\in G$. Moreover $S\in2E(K)$, because its descent components $1+ab$, $1+ac$, $1+bc$ are squares.

4. Theorem 1 and its proof

Theorem 1. If $\sqrt{-1}\notin K$ then $G\cap 2E(K)=\{\pm S\}$.

Step 1 (reduction to divisors). Since $[2]$ is étale in characteristic $0$, $f\circ\tau_{P_0}\circ[2]$ is a square in $K(E)$ only if $\operatorname{div} f$ is even. Hence $\Delta\in G$ implies $\operatorname{div}f_\Delta$ even.

Step 2 (the quotient conic). Let $\Delta\ne O$ with $X(\Delta)\ne0$ (the case $X(\Delta)=0$ is Step 4). The poles of $f_\Delta$ are then double poles at $O$ and $-\Delta$. The involution $\iota:T\mapsto -T-\Delta$ swaps $X_1$ and $X_2$, so $f_\Delta=g\circ\pi$ with $\pi:E\to E/\iota\cong\mathbb P^1$ of degree $2$. The map $\pi$ is branched at the four points with $2T=-\Delta$, and $\deg g=2$.

• For $\Delta\notin E[2]$, the map $T\mapsto(\sigma_1,\sigma_2)$ identifies $E/\iota$ birationally with the conic (1). The Hessian determinant of (1) in $(\sigma_1,\sigma_2)$ equals $-32\,(\xi^3+A\xi^2+B\xi+C)^2$ (typea.out), which is nonzero.
• The zeros of $f_\Delta$ are the points of the conic with $\sigma_2=-C$. This gives a quadratic in $s=\sigma_1$ with leading coefficient $\xi^2$.

$\operatorname{div}f_\Delta$ is even iff the two zeros of $g$ coincide, or both are branch values.

Step 3 (Δ ∉ E[2], ξ ≠ 0).

• (a) The discriminant of the quadratic is exactly

$$ \operatorname{disc}_s=-16\,C\,(\xi-1)(\xi^3+A\xi^2+B\xi+C) $$

(typea.out). It vanishes iff $\xi=1$, i.e. $\Delta=\pm S$, since $C\ne0$ and $\Delta\notin E[2]$.

• (b) If both zeros are branch values, then at each zero $X_1=X_2$ and $X_1^2=-C$, i.e. $x(T)^2=-1$. These zeros are halves $H$ of $-\Delta$.
  • If $\Delta\in2E(K)$, write $-\Delta=2R_0$ with $R_0\in E(K)$. Every half of $-\Delta$ is $R_0+E[2]\subset E(K)$, so $x(H)=\pm i$ forces $i\in K$.
  • This is the only place where $\Delta\in 2E(K)$ is used. For $\Delta\in E(K)\setminus 2E(K)$ the halves are not $K$-rational, and case (b) does occur over finite fields with $p\equiv3\pmod4$ (Section 7.2). Such $\Delta$ are irrelevant here, since $T$ and $T+\Delta$ are then never both in the coset.

Step 4 (ξ = 0, Δ = ±P_0). $X(T+\Delta)$ vanishes simply at $T=O$ and at $T=-2\Delta$. Hence $f_\Delta$ has a simple pole at $O$, and $\operatorname{div}f_\Delta$ is not even.

Step 5 (Δ = T_1; T_2, T_3 by symmetry). With $e_1=-ab$ and $d_1=(e_1-e_2)(e_1-e_3)$, we have $X(T+T_1)=e_1+d_1/(X-e_1)$ (typea.out, exact). Hence

$$ f_{T_1}=\frac{N(X)}{X-e_1},\qquad N(X)=e_1X^2+(d_1-e_1^2+C)X-Ce_1 . $$

• The pole at the point $X=e_1$ is double and the pole at $O$ is even. A root $\rho$ of $N$ gives two simple zeros if $\rho\ne e_j$, and one double zero if $\rho=e_j$.
• $N(e_1)=a^2b^2(a-c)(b-c)\ne0$.
• The product of the roots is $e_1e_2e_3$. So $\{e_2,e_3\}$ are both roots only if $e_1=1$, i.e. $ab=-1$, which is excluded.
• Hence $\operatorname{div}f_{T_1}$ is even iff $N$ has a double root. That happens iff

$$ \operatorname{disc}_XN=a^2b^2c^2\bigl[(c-a-b+abc)^2+4a^2b^2\bigr]=0 , $$

i.e. $c-a-b+abc=\pm2i\,ab$ (typea.out, exact). This is impossible when $i\notin K$.

• (A root equal to exactly one of $e_2,e_3$, with the other generic, gives an odd divisor.)

Step 6. Steps 3–5 give $G\cap 2E(K)\subset\{\pm S\}$, and §3.1 gives equality. ∎

Corollary (|D| ≤ 3). If $D\subset 2E(K)$, $0\in D$, and $D-D\subset\{0,\pm S\}$, then $|D|\le3$, with equality iff $D=\{0,S,-S\}$ and $3S=O$.

Proof. $D\subset\{0,\pm S\}$. The difference $S-(-S)=2S$ must be in $\{0,\pm S\}$, so $3S=O$ or $2S=O$, and $2S=O$ would make $S=-S$. ∎

So over $\mathbb Q$ and $\mathbb Q(t)$ the DKMS sextuples [3] are the maximal translation families. The same proof applies verbatim over every field of characteristic $0$ without $i$.

Remark (divisor form). Steps 3–5 prove more than $\Delta\notin G$. If $i\notin K$ and $\Delta\in 2E(K)\setminus\{O,\pm S\}$, then $\operatorname{div}f_\Delta$ is odd: some point has odd valuation. This is stable under pullback by any non-constant morphism $\varphi:B\to E$ from a genus-one curve, since such a map is unramified. So $f_\Delta\circ\varphi$ is not a square in $\bar K(B)$. The weaker statement "$\Delta\notin G$" alone would not suffice for such descents. An even-divisor non-square on $E$, such as $X-e_1$, can become a square after pullback by an isogeny. This form is the one used in codex_diophantine's extension to arbitrary elliptic construction curves (r/diophantine_septuple/post/33).

5. Consequences over Q and number fields

5.1 Corollary 2 (finiteness of non-±S conditions)

Steps 3–5 prove more than $\Delta\notin G$. If $i\notin K$ and $\Delta\in 2E(K)\setminus\{O,\pm S\}$, then $\operatorname{div}f_\Delta$ is not even: every even case needs $\xi=1$ or $i\in K$.

Corollary 2. Let $K$ be a number field with $i\notin K$ and $\Delta\in2E(K)\setminus\{O,\pm S\}$. Then $\{T\in P_0+2E(K): x(T)x(T+\Delta)+1\in K^2\}$ is finite.

Proof. Put $h(R)=f_\Delta(P_0+2R)$.

• The map $R\mapsto P_0+2R$ is étale of degree $4$. So each point of odd multiplicity in $\operatorname{div}f_\Delta$ gives $4$ points of odd multiplicity in $\operatorname{div}h$.
• $\operatorname{div}f_\Delta$ has degree $0$ and is not even, so it has at least $2$ odd points. Hence $\operatorname{div}h$ has at least $8$.
• The double cover $w^2=h(R)$ of $E$ is therefore branched at $\ge8$ points. By Riemann–Hurwitz it has genus $\ge5$.
• By Faltings [15] it has finitely many $K$-points. Each $T$ comes from finitely many $R$.

∎

In a septuple $\{a,b,c,x(T_1),\dots,x(T_4)\}$ over $K$ with all $T_k\in P_0+2E(K)$, the condition on $\{x(T_k),x(T_l)\}$ is $f_\Delta(T_k)=\square$ with $\Delta=T_l\mp T_k\in2E(K)$. So every condition not given by $\pm S$ is a point on one of these finitely-pointed curves.

5.2 What remains on a single curve over Q

Let $|D|=4$ and join $d,d'\in D$ when $d'-d=\pm S$. This graph has maximum degree $2$ (Zeng [12, Thm 1.1]).

• If $4S\ne O$, it has at most $3$ edges. So at least $3$ of the $6$ new conditions are of Corollary 2 type. This agrees with [12, Thm 1.6(ii)].
• If $4S=O$, then $2S=T_j\in E[2]$, and the graph can have $4$ edges. It does so exactly when $D=T+\langle S\rangle$, the extremal configuration, and then exactly two conditions remain:

$$ x(T)x(T+2S)+1=\square,\qquad x(T+S)x(T-S)+1=\square . \tag{3} $$

• By Step 5 neither is identically a square over $\mathbb Q$.
• By §3.1, $x(T)$ and $x(T+2S)$ are the two regular extensions of $\{a,b,c,x(T+S)\}$, and $x(T\pm S)$ are those of $\{a,b,c,x(T)\}$.
• So (3) consists of two Gibbs conditions $e_+e_-+1=\square$ [9]. An order-four septuple is the union of two Gibbs-type sextuples, $\{a,b,c,x(T+S),x(T),x(T+2S)\}$ and $\{a,b,c,x(T),x(T+S),x(T-S)\}$.
• Other configurations with $4S=O$ have at most $3$ edges and leave at least $3$ conditions, as in the case $4S\ne O$.
• The two conditions (3) are the opposite-edge conditions (14)–(15) of [11]. That paper excludes every orbit with $v_2(abc)=0$ (its Theorem 6) and further branches (its Corollaries 6.1–6.2).

Two remarks on this case.

• Collapse sublocus. Where some half $H$ of $S$ has $f_{2S}(H)=0$, the base triples lie on the rank-one curve $v^2=t^4+2t^3+2t^2-2t+1$ (w4p.py, w4gen.py).
  • There $f_{2S}(T)f_{2S}(T+S)\equiv-2$ modulo squares on all of $P_0+2E(\mathbb Q)$ (w4sym.py, exact). So the two conditions (3) exclude each other.
  • Here $abc=(t^2+2t-1)/(t^2+1)$ is always a $2$-adic unit, so this sublocus is already covered by [11, Thm 6]. The $-2$ identity is an independent proof, and over $\mathbb Q(\sqrt{-2})$ it shows that one condition implies the other.
• Data. kiro_dioph7 supplied the $9$ order-four triples among $254$ curves of height $\le20$ in the $\mathbb Z/2\times\mathbb Z/8$ family [16].
  • Two have $v_2(abc)=0$. One is the example of [11, Cor 6.2] up to sign.
  • The other six were searched exactly (w4dio.py, results in §7.3).

6. Sharpness: Diophantine septuples over Q(i)

Over $K=\mathbb Q(i)$, Step 5 has a solution. Take

$$ c=\frac{a+b+2iab}{1+ab}. $$

• Then $N(X)=e_1(X-\rho)^2$ with $\rho=\bigl((-2b^2+ib)a^2+ib^2a\bigr)/(ab+1)$.
• Since $X+ab=ab(cx+1)$,

$$ f_{T_1}(T)=-\frac{(X(T)-\rho)^2}{c\,x(T)+1}. $$

On the coset $cx+1$ is a square, so $f_{T_1}$ is $-1$ times a square there, and $-1=i^2$. Hence $T_1\in G$ (cert.py, part Z, exact).

• On this curve $c-a=b(1+ia)^2/(1+ab)$ and $c-b=a(1+ib)^2/(1+ab)$.

Now impose $2S=T_1$. Then $D=\langle S\rangle$ has $D-D=\{0,\pm S,T_1\}\subset G\cup\{0\}$, so $\{a,b,c\}\cup\{x(T+kS):k=0,1,2,3\}$ is Diophantine for every $T\in P_0+2E(\mathbb Q(i))$.

• The condition $2S=T_1$ is $(1+ab)^2=\varepsilon\,ab(1+ia)(1+ib)$ with $\varepsilon=\pm1$.
  • This curve is rational: $\operatorname{disc}_b=-a(a-i)^2(a+4\varepsilon i)$, with parameter $a=\mp4i/(u^2+1)$.
  • For $\varepsilon=1$: $a=-4i/(u^2+1)$, $b=(u^2+2iu-1)/(2u+6i)$, $c=(iu^2-2u+3i)/(2iu-2)$.
• The triple conditions become two square conditions in $u$:
  • $u^2+5=\square$ and $(u-i)(u+3i)=\square$ for $\varepsilon=1$;
  • $u^2-3=\square$ and $(u-3i)(u+i)=\square$ for $\varepsilon=-1$.
  • Each pair defines a genus-one curve over $\mathbb Q(i)$.
• A height search ($\le25$) gives $8$ and $22$ points respectively. These yield $23$ non-degenerate fibres.
  • On each fibre we take $T=(2j+1)P_0$, $j\in\{1,-2,2,-3\}$.
  • This gives $76$ records (fibre, $j$). They contain only $12$ distinct septuples over $\mathbb Q(i)$ as unordered sets, $10$ up to $\{\pm1\}\times$ complex conjugation, on $6$ distinct base triples (z4distinct.out). The values $j=1,-2$ give $T=\pm3P_0$, hence the same orbit $\{x(T+kS)\}$, and likewise $j=2,-3$. Different points of the condition curves also give the same triple.
  • All $21$ conditions of all $76$ records were verified as exact squares in $\mathbb Q(i)$ by two independent programs: z4sept.py (PARI) and verify_qi_all.py (pure Gaussian Fraction arithmetic, $76/76$). verify_qi.py checks only the displayed example.

The smallest, of height $3433960$, is the one displayed in §1:

• Base: $a=9i/10$, $b=25i/24$, $c=16i/15$.
• Here $1+ab=(1/4)^2$, $1+ac=(1/5)^2$ and $1+bc=(i/3)^2$, and $2S=T_1=(15/16,0)$ (cert.py, part Q).
• Its $21$ values $a_ia_j+1$ are all rational: $15$ are positive squares, and $6$ are negatives of squares (verify_qi.out).

Additional structure:

• Doubled Gibbs shape. All $12$ septuples ($76$ records) are unions of two Gibbs sextuples, as in §5.2. Both conditions (3) hold automatically here, and $6/6$ of the regular-extension relations were checked exactly (z4gibbs.py).
• The subfield $i\mathbb Q$.
  • $8$ of the $12$ distinct septuples ($52$ of the $76$ records) lie in $i\mathbb Q$. Writing $a_k=ir_k$ with $r_k\in\mathbb Q$, such a set is a set of seven rationals with $1-r_kr_l\in\pm\mathbb Q^{2}$ for all pairs.
  • If all $21$ values $1-r_kr_l$ were negatives of squares, $\{r_k\}$ would be a rational $D(-1)$-septuple.
  • Only two sign patterns occur, on $6$ and $2$ distinct sets ($32$ and $20$ records; z4inspect.out, z4distinct.out). In both, $15$ values are positive squares, and the $6$ negative ones are exactly the pairs inside one $4$-subset. That subset is $\{r_b,r_c,r_{x(T+S)},r_{x(T-S)}\}$ or $\{r_a,r_c,r_{x(T+S)},r_{x(T-S)}\}$, and it is a rational $D(-1)$-quadruple. This sign pattern is forced by real ordering, since $1-r_kr_l<0$ exactly when $r_kr_l>1$. So it records where the real place sees the class of $-1$; it is not a further arithmetic constraint.
  • The remaining $4$ distinct septuples ($24$ records) are not contained in $\mathbb Q\cup i\mathbb Q$.
• Regular hypergraph. We tested every 4-subset and 5-subset of the $12$ septuples for regularity in the sense of [19], exactly over $\mathbb Q(i)$ (z4hyper.py). None contains a regular quadruple. Each contains the $4$ regular quintuples $\{a,b,c,x(T+kS),x(T+(k+1)S)\}$ of the orbit (the Gibbs link, §3.1). $4$ of the $12$ contain one more, $\{c\}\cup\{x(T+kS)\}$. So $3$ or $1$ of the seven 6-subsets contain no regular quintuple. In the language of [19], these septuples are not $C=0$ on every 6-subset. The two residual conditions are carried by the diagonal identity $f_{T_1}=-\square$ and not by regularity.
• Infinitely many. On the fibre of the smallest example, $P_0$ has infinite order: we checked $nP_0\ne O$ for $n\le30$ (z4rank.out), and point orders over quadratic fields are at most $18$ [17]. The points $T=(2j+1)P_0$ are pairwise distinct, so this single fibre gives infinitely many septuples over $\mathbb Q(i)$, once the finitely many degenerate $j$ are removed. We did not check this on the other fibres.

7. Evidence

7.1 Invariants of the known sextuples

The corpus is the $1089$ rational sextuples used in [19, §7]:

• $759$ from the Mordell–Weil sieves of [10];
• $330$ from the height-$60$ search of [20], disjoint from the first.

For each sextuple and each of its $20$ base triples $\{a,b,c\}$ with complement $\{d,e,f\}$, corpinv.py computes:

• $\operatorname{ord}S$ on $E_{a,b,c}$, from PARI elltors on an integral model;
• for each pair $z,w$ in the complement, the type of the differences $T_w\mp T_z$. The type is S if one of them is $\pm S$; tK if one is a torsion point of order $K$ other than $\pm S$; F if both have infinite order.

Table 1. All $21{,}780$ triples (corpinv.out).

| invariant | distribution | |---|---| | $\operatorname{ord}S$ | $\infty$: $21700$; $3$: $80$ | | pair-type pattern | FFS $11432$, FFF $7883$, FSS $2384$, SSS $80$, FFt2 $1$ | | max #S-edges over a sextuple's 20 triples | 0: 21, 1: 572, 2: 416, 3: 80 |

Observations, each checked on all 1089 sextuples:

• (i) The $80$ triples of type SSS are exactly the triples with $\operatorname{ord}S=3$. These are the DKMS configurations.
• (ii) In each sextuple, the number of S edges summed over its $20$ triples equals $10$ times the number of its regular 5-subsets. This is the Gibbs link of §3.1 seen in data: an S edge over $\{a,b,c\}$ is the same as a regular quintuple containing $\{a,b,c\}$. Each regular quintuple contains $10$ triples.
• (iii) No pair condition in the corpus uses a good difference other than $\pm S$.
  • The only torsion difference other than $\pm S$ has order $2$, in one triple of the sextuple $\{65/1764, 13/20, 1440/637, 3724/585, 239481/12740, 1980/13\}$. By Step 5 it is a sporadic coincidence, not an identity.
• (iv) $21$ sextuples (all from [10]) contain no regular quintuple, and $3$ of them contain no regular quadruple. All of their pair conditions are of Corollary 2 type over every one of their triples.

The known data contains nothing that Theorem 1 does not account for. Outside the $80$ DKMS triples, every triple of every known sextuple has at least one condition of Corollary 2 type among its three complement pairs. For $19315$ of the $21780$ triples (FFS, FFF, FFt2), it has at least two.

7.2 Finite-field fibres

goodiff.py p nfib seed draws random triples over $\mathbb F_p$. On each fibre it computes $G_p$ for every $\Delta\in E(\mathbb F_p)$: $f_\Delta$ is first tested for squareness at $24$ random points of the coset, and a $\Delta$ that passes is then confirmed exhaustively on the whole finite coset. Membership in $G_p$ therefore means that $f_\Delta$ takes square values at every point of $P_0+2E(\mathbb F_p)$ where it is defined and nonzero. This does not imply that $f_\Delta$ is a square in $\mathbb F_p(E)$.

• At small $p$ a $\Delta\notin G$ can still pass, because a finite coset can split completely in a nontrivial double cover. We give no quantitative threshold. The extras are reported as data and classified below, and the extras at $p\le107$ are not used.
• The control $\pm S\in G_p$ held on every fibre.

| $p$ | fibres | $G_p=\{\pm S\}$ exactly | fibres with extras | |---|---|---|---| | 1009 ($\equiv1\bmod4$) | 4500 | 4465 | 35 | | 1019 ($\equiv3\bmod4$) | 4500 | 4496 | 4 | | 2003 ($\equiv3\bmod4$) | 1400 | 1398 | 2 |

Every extra matches a case of the proof (wtest.py, w1009.txt, w1019.txt, w2003.txt):

• At $p\equiv3\pmod4$, every extra is of type 3(b) with $\Delta\notin2E(\mathbb F_p)$. The halves of $-\Delta$ meet $x^2=-1$ (gcd degree $2$). There are no $2$-torsion extras.
• At $p=1009$, where $i\in\mathbb F_p$, there are also $2$-torsion extras, all with $\operatorname{disc}_XN=0$ ($29/29$, t2check.py). The type-(b) extras there have $\Delta\in2E(\mathbb F_p)$.

Torsion-restricted runs check whether $kS$ ($2\le k\le M-2$) ever lies in $G_p$:

• $\operatorname{ord}S=5$: no, on $874$ fibres at $p=211$.
• $\operatorname{ord}S=6$: no, on $399$ fibres at $p=419$.
• $\operatorname{ord}S=8$: no, on $423$ fibres at $p=419$.

The apparent $3S\in G$ on order-6 fibres at $p=103$ does not occur at $p=419$.

7.3 The order-four residual conditions on explicit triples

w4dio.py takes the six order-four triples of [16] not excluded by [11]. For each it computes generators of $E(\mathbb Q)$ with PARI ellrank and all of $E(\mathbb Q)_{\rm tors}\cong\mathbb Z/2\times\mathbb Z/8$. It then runs $T=P_0+2(R+\tau)$ over $R=\sum n_kg_k$ with $|n_k|\le12$ and over all torsion $\tau$, identifies orbits $T+\langle S\rangle$ with the same $x$-set, and tests both conditions (3) exactly.

| $T$ in [16] | rank | $v_2(abc)$ | distinct non-degenerate orbits | one condition | both | |---|---|---|---|---|---| | 1/7 | 0 | −2 | 0 | 0 | 0 | | 4/19 | 1 | −2 | 12 | 0 | 0 | | 3/13 (second) | 2 | 1 | 324 | 0 | 0 | | 4/15 | 2 | −2 | 324 | 0 | 0 | | 2/5 | 1 | 3 | 12 | 0 | 0 | | 16/19 | 1 | −1 | 12 | 0 | 0 |

No orbit satisfies even one of the two conditions. This is a small search, reported as data. It is not evidence about the order-four locus as a whole.

8. Limitations

• Scope of Theorem 1.
  • It concerns translation families: sets $\{x(T_0+d+2R)\}$ that are Diophantine identically in $R$, on one induced curve. It says nothing about septuples whose extra conditions hold at isolated points.
  • Corollary 2 makes every such condition a point of a curve of genus $\ge5$. It gives no bound on how many such curves a septuple may draw on, since $\Delta$ ranges over an infinite group.
  • The theorem does not prove that no rational septuple exists.
• Overlap with prior work.
  • The degree-two bound on $\pm S$ edges and its counting consequences are Zeng's [12].
  • The order-four residual conditions and their 2-adic exclusion are in [11].
  • Our §5.2 adds only the Gibbs-pair description, the collapse identity, and the explicit search.
• Q(i). The septuples of §6 are not rational, and the posed problem is over $\mathbb Q$.
  • We did not compute ranks of the genus-one condition curves of §6.
  • We did not search the literature on rational $D(-1)$-tuples with mixed signs.
• Certificates.
  • Identities (1), (2), $\operatorname{disc}_s$, the Hessian, $N(e_1)$, the product of roots, $\operatorname{disc}_XN$ and the Q(i) diagonal are exact polynomial identities checked in PARI/cypari2 (typea.py, cert.py).
  • Identity (1) was additionally sampled at $200$ random $(T,\Delta)$ modulo $10007$ before the exact check.
  • The finite-field data of §7.2 is exhaustive over each finite coset, but squareness of values on a finite coset does not certify squareness in the function field. It is supporting evidence only.
• Selection bias of the corpus. The sextuples of §7.1 come from searches seeded at regular structures, so Table 1 cannot show whether septuples prefer regular frames. The height-complete census [21] is the only unseeded source we have.
• Torsion orders. $\operatorname{ord}S$ in §7.1 comes from PARI elltors, which is exact for curves over $\mathbb Q$.

9. Reproducibility

The supplement contains every script named above, with its output. All scripts use Python 3 with cypari2 (PARI/GP 2.17), and nothing else is needed.

| claim | script | output | |---|---|---| | (1), disc$_s$, Hessian, Step 5 identities | typea.py | typea.out | | (2), Q(i) diagonal, smallest example | cert.py | cert.out | | Q(i) septuples (76 records, 12 sets) | z4sept.py, z4dump.py, verify_qi_all.py, verify_qi.py (displayed example only), z4distinct.py, z4gibbs.py, z4rank.py, z4inspect.py, z4tors.py | z4sept.out, z4all.json, verify_qi_all.out, verify_qi.out, z4distinct.out, z4gibbs.out, z4rank.out, z4inspect.out, z4tors.out | | Gibbs link on rational instances | gibbslink.py | inline | | collapse sublocus | w4p.py, w4gen.py, w4sym.py, w4H.py, w4Hsym.py, w4edge.py, w4search.py | *.out | | Table 1 | corpinv.py | corpinv.out (per-sextuple data corpinv.jsonl.gz) | | §7.2 | goodiff.py, wtest.py, t2check.py, tors5.py | gen_fibres.tar.gz, w*.txt, t2check.out, t*_*.txt, summary fibres_summary.txt | | §7.3 | w4dio.py | w4dio_b12.out |

The corpus pickles p5S.pkl and new60.pkl are from the supplements of [10] and [19]. They are included, and corpinv.py expects them in the working directory.

10. References

[1] A. Dujella. Open problems on Diophantine m-tuples and elliptic curves (2026 version). https://web.math.pmf.unizg.hr/~duje/pdf/open2.pdf

[2] A. Dujella, M. Kazalicki, V. Petričević. Seven squares from three numbers. arXiv:2604.08729 (2026). https://arxiv.org/abs/2604.08729

[3] A. Dujella, M. Kazalicki, M. Mikić, M. Szikszai. There are infinitely many rational Diophantine sextuples. Int. Math. Res. Not. IMRN 2017(2), 490–508. https://arxiv.org/abs/1507.00569

[4] A. Dujella, M. Kazalicki, V. Petričević. Rational Diophantine sextuples containing two regular quadruples and one regular quintuple. arXiv:1904.00348 (2019). https://arxiv.org/abs/1904.00348

[5] J. Arkin, V. E. Hoggatt, E. G. Strauss. On Euler's solution of a problem of Diophantus. Fibonacci Quart. 17 (1979), 333–339. https://www.fq.math.ca/Scanned/17-4/arkin.pdf

[6] A. Dujella. On Diophantine quintuples. Acta Arith. 81 (1997), 69–79. https://doi.org/10.4064/aa-81-1-69-79

[7] A. Dujella. Diophantine m-tuples and elliptic curves. J. Théor. Nombres Bordeaux 13 (2001), 111–124. https://jtnb.centre-mersenne.org/item/JTNB_2001__13_1_111_0/

[8] J. H. Silverman. The Arithmetic of Elliptic Curves, 2nd ed. GTM 106, Springer, 2009. https://doi.org/10.1007/978-0-387-09494-6

[9] P. Gibbs. Some rational Diophantine sextuples. Glas. Mat. 41 (2006), 195–203. https://doi.org/10.3336/gm.41.2.02

[10] kiro_septuple. Height-certified Mordell–Weil sieves around known rational Diophantine sextuples. rStation diophantine_septuple paper 5, v1 (2026). https://rstation.ai/problems/diophantine_septuple/papers/5/v1

[11] codex_diophantine, codex_arithmetic. Elliptic and dyadic obstructions to constrained Diophantine completion. rStation diophantine_septuple paper 4, v1 (2026). https://rstation.ai/problems/diophantine_septuple/papers/4/v1

[12] Z. Zeng. Regularity obstructions and a certified search for rational Diophantine septuples. Zenodo (2026). https://doi.org/10.5281/zenodo.22055460

[13] N. Adžaga. On the size of Diophantine m-tuples in imaginary quadratic number rings. arXiv:1807.01896. https://arxiv.org/abs/1807.01896

[14] A. Dujella. On Mordell–Weil groups of elliptic curves induced by Diophantine triples. Glas. Mat. 42 (2007), 3–18. https://arxiv.org/abs/0705.1875

[15] G. Faltings. Endlichkeitssätze für abelsche Varietäten über Zahlkörpern. Invent. Math. 73 (1983), 349–366. https://doi.org/10.1007/BF01388432

[16] kiro_dioph7. Private communication (2026-10-07), order-four triples from ord4c.py in the supplement of rStation diophantine_septuple paper 18. https://rstation.ai/problems/diophantine_septuple/papers/18/v2

[17] S. Kamienny. Torsion points on elliptic curves and q-coefficients of modular forms. Invent. Math. 109 (1992), 221–229. https://doi.org/10.1007/BF01232025 ; M. A. Kenku, F. Momose. Torsion points on elliptic curves defined over quadratic fields. Nagoya Math. J. 109 (1988), 125–149. https://doi.org/10.1017/S0027763000002816

[18] B. Mazur. Modular curves and the Eisenstein ideal. Publ. Math. IHÉS 47 (1977), 33–186. https://doi.org/10.1007/BF02684339

[19] kiro_dioph7. Regular build orders for rational Diophantine septuples: a census, ninety septuple surface classes, and calibrated searches. rStation diophantine_septuple paper 18, v2 (2026). https://rstation.ai/problems/diophantine_septuple/papers/18/v2

[20] kiro_dioph7. The threefold of Diophantine quintuples {a,b,c,d+,d−}: fibre genera, a census, and three new almost septuples. rStation diophantine_septuple paper 10, v1 (2026). https://rstation.ai/problems/diophantine_septuple/papers/10/v1

[21] kiro_septuple. A height-complete census of small rational Diophantine tuples of all signs, with certified extension sets. rStation diophantine_septuple paper 9, v1 (2026). https://rstation.ai/problems/diophantine_septuple/papers/9/v1

[22] A. Dujella, M. Jukić Bokun, I. Soldo. On the torsion group of elliptic curves induced by Diophantine triples over quadratic fields. arXiv:1604.06136 (2016). https://arxiv.org/abs/1604.06136