Zilber–Pink: when too many special intersections force structure

← All digests

Family 016 · Number theory
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-10

1. The question

Two surfaces in three-dimensional space usually meet along a curve. Two curves in that space usually miss each other. Algebraic geometry packages this expectation in the formula

$$\text{expected dimension}=\dim X+\dim T-\dim S,$$

where $X,T$ lie in an ambient space $S$. An intersection larger than this prediction suggests a structural reason.

The Zilber–Pink principle asks what happens when $T$ ranges over infinitely many arithmetically special subvarieties. A fixed pair might be easy to study, but infinitely many moving special targets could conceivably produce infinitely many exceptional intersections. The conjecture predicts strong finiteness once the special structure already forced by $X$ is taken into account.

Family 016 contains four manuscripts. M1 claims the general abelian version over algebraic numbers. M2–M4 treat three types of special points on curves in the moduli space of principally polarized abelian surfaces, culminating in the curve case there. The two settings share a philosophy but have different meanings of “special.”

2. Background and essential definitions

The fixed abelian variety

An abelian variety is a connected projective algebraic group. An elliptic curve is the one-dimensional example; a product of elliptic curves gives examples in higher dimensions. A torsion point $\tau$ satisfies $m\tau=0$ for some positive integer $m$.

Inside an abelian variety $A$, a special subvariety is a torsion coset

$$T=B+\tau,$$

where $B$ is a connected abelian subgroup. This combines a subgroup direction with a finite-order translation. For an irreducible subvariety $X$, let $S=\langle X\rangle_{\mathrm{sp}}$ be the smallest torsion coset containing it.

An irreducible component $Y$ of $X\cap T$, with $T\subseteq S$ special, is atypical if

$$\dim Y>\dim X+\dim T-\dim S.$$

The ambient dimension is that of $S$, not automatically that of $A$. Otherwise a subvariety already lying in a smaller torsion coset would create misleading exceptional intersections.

For example, a curve in a three-dimensional abelian variety and a special curve have expected intersection dimension $-1$. Any intersection point is therefore atypical. A special surface has expected intersection dimension zero with the same curve, so an isolated intersection point is not atypical by this test.

“Maximal atypical” means maximal by inclusion among atypical subvarieties of $X$. Finiteness of maximal ones allows an infinite collection of exceptional points to lie inside one positive-dimensional exceptional subvariety.

The varying abelian surface

The moduli space $\mathcal A_2$ parametrizes principally polarized abelian surfaces. A polarization supplies the algebraic analogue of a positive form; principal means its associated map to the dual abelian variety is an isomorphism. The space $\mathcal A_2$ has dimension three, so it is often called the Siegel threefold.

Its special subvarieties are Shimura-special loci, including their Hecke translates. They encode extra endomorphisms or related Hodge-theoretic structure. A Hecke translate changes the abelian variety by a suitably compatible finite isogeny. An isogeny is a surjective homomorphism with finite kernel. A curve is Hodge generic if it lies in no proper special subvariety.

For an elliptic curve, complex multiplication, or CM, means that its geometric rational endomorphism algebra is an imaginary quadratic field instead of $\mathbb Q$. For an abelian surface $A_s$, write

$$\operatorname{End}^0(A_s)=\operatorname{End}_{\overline{\mathbb Q}}(A_s)\otimes_{\mathbb Z}\mathbb Q.$$

The full algebra matters. The matrix algebra $M_2(\mathbb Q)$ corresponds to a surface isogenous to the square of a non-CM elliptic curve. An indefinite quaternion division algebra is instead a four-dimensional noncommutative algebra over $\mathbb Q$, with every nonzero element invertible, which becomes $M_2(\mathbb R)$ over $\mathbb R$. It is not the split algebra $M_2(\mathbb Q)$.

3. Development history and previous partial results

Raynaud’s Manin–Mumford theorem established the torsion-point case: an irreducible subvariety with Zariski-dense torsion points must itself be a torsion coset. Zilber’s unlikely-intersection formulation and Pink’s semiabelian and Shimura formulations extend the problem to moving special subvarieties of positive dimension.

In abelian varieties, Viada and Rémond–Viada obtained early curve results in powers of CM elliptic curves. Habegger and Pila proved the curve theorem for arbitrary abelian varieties over number fields. Their 2016 paper also gives higher-dimensional results under additional quotient-dimension or height restrictions. Barroero, Kühne and Schmidt later treated curves in general semiabelian varieties. M1’s introduction distinguishes these curve theorems from unrestricted higher-dimensional maximal-atypical finiteness.

Higher-dimensional partial results included non-density when every quotient has the expected image dimension, non-density at a fixed height bound, and finiteness of certain codimension-one or codimension-two exceptional subvarieties. These are meaningful advances, but a fixed height cutoff cannot cover points of arbitrarily large height without another argument.

For $\mathcal A_2$, André–Oort supplies finiteness of special points on a Hodge-generic curve. Pila and Tsimerman established this case, and Tsimerman’s later general theorem covers the broader Siegel setting. Positive-dimensional special targets require more.

Daw and Orr developed reductions from large Galois orbits to finiteness and proved important cases under degeneration assumptions. Their work on large Galois orbits under multiplicative degeneration obtains the full curve conclusion when the closure meets the zero-dimensional boundary stratum. Papas obtained further interior-period results with hypotheses involving reduction and the number of supersingular places of proximity. M2–M4 aim to remove such restrictions for the three relevant loci.

The common counting strategy comes from Pila and Zannier: arithmetic supplies many conjugates, while tame geometric counting allows few unless an unexpected algebraic structure appears. The new manuscripts place their height and period estimates within that existing framework.

4. What the new manuscripts claim

M1: finitely many maximal atypical subvarieties

Let $A$ be any abelian variety over a number field and $X\subseteq A_{\overline{\mathbb Q}}$ irreducible. M1 claims that $X$ has only finitely many maximal atypical subvarieties, measured inside $\langle X\rangle_{\mathrm{sp}}$.

No restriction on dimension, simplicity, complex multiplication, point height or point degree is imposed. A key intermediate statement is universal non-density. If $X$ lies in no proper torsion coset and $d=\dim X$, then

$$X\cap A^{[d+1]} \quad\text{is not Zariski dense in }X,$$

where $A^{[r]}$ is the union of all torsion cosets of codimension at least $r$. Zariski non-density means containment in a proper algebraic closed subset. For an irreducible curve, a proper closed subset is finite; in higher dimensions, non-density alone is weaker than finiteness.

M1 applies a universal reduction through optimal subvarieties to pass from this non-density theorem for all ambient varieties to the maximal-atypical conclusion. It also claims a finite-rank translated version: if $\Gamma\subset A(\overline{\mathbb Q})$ satisfies

$$\dim_{\mathbb Q}(\Gamma\otimes_{\mathbb Z}\mathbb Q)<\infty,$$

and $X$ lies in no translate of a proper algebraic subgroup, then

$$X(\overline{\mathbb Q})\cap\bigl(A^{[d+1]}(\overline{\mathbb Q})+\Gamma\bigr)$$

is not Zariski dense. This stronger geometric hypothesis excludes arbitrary proper translates, not just torsion translates. The group $\Gamma$ need not be finitely generated: division hulls and torsion groups are included.

M2–M4: three special-curve branches

Let $C\subset\mathcal A_2$ be a reduced irreducible closed Hodge-generic curve over $\overline{\mathbb Q}$. The manuscripts claim finiteness for the following sets of algebraic points:

Manuscript	Condition on the represented abelian surface
M2	Isogenous to $E_1\times E_2$, with at least one elliptic factor having CM.
M3	Full geometric rational endomorphism algebra equal to an indefinite quaternion division algebra over $\mathbb Q$.
M4	Isogenous to $E^2$ for a non-CM elliptic curve $E$, equivalently full endomorphism algebra $M_2(\mathbb Q)$.

The isogenies in the product descriptions need not preserve the given principal polarization. All degrees, CM fields, endomorphism orders, relevant quaternion algebras and polarizations are allowed. None of these statements requires the curve to meet a boundary, and complete curves are included.

M4 combines the three branches with the classification of special curves and André–Oort for special points. Its resulting theorem is

$$C(\overline{\mathbb Q})\cap \bigcup_{\substack{Z\subset\mathcal A_2\ \mathrm{special}\\\dim Z\le1}} Z(\overline{\mathbb Q}) \quad\text{is finite}.$$

For a curve not Hodge generic in $\mathcal A_2$, M4 also records the corresponding formulation inside its smallest special closure $S_C$, using special targets with $\dim Z<\dim S_C-1$.

A quantitative ingredient, not an effective list

For a mixed product point in M2, let $N(s)$ be the least product-isogeny degree and $D$ the absolute discriminant of the CM factor’s endomorphism order. Set $c(s)=\max\{N(s),D\}$. M2 claims

$$\#\bigl(\operatorname{Gal}(\overline{\mathbb Q}/K)\cdot s\bigr) \ge c\,c(s)^{1/42},$$

with $c>0$ depending on the fixed curve and its field of definition $K$. The exponent is a usable positive power, not an optimized constant. The statement does not claim effectivity; its inputs include a potentially ineffective class-number estimate.

5. Why the advance matters

Provisional importance: 5/5. Taken as stated, M1 settles a central arbitrary-dimensional form of the abelian Zilber–Pink conjecture. M2–M4 complete another substantial case, curves in $\mathcal A_2$, by treating all three special-curve branches without boundary restrictions.

The broad idea is that unexpectedly large intersections cannot occur in infinitely many unrelated ways. In the abelian setting they must be contained in finitely many maximal exceptional subvarieties. In the Hodge-generic curve setting the exceptional set itself is finite.

These conclusions strengthen the structural meaning of a dimension count. They also connect diverse methods: canonical heights, period estimates, reduction at primes, Galois orbits and geometry of definable sets. The scope merits the highest relative rating without implying that all mixed-Shimura or semiabelian versions are covered.

6. The main idea

Separate the height scales

M1 starts from a hypothetical Zariski-generic sequence of unlikely-intersection points. The canonical, or Néron–Tate, height behaves quadratically under multiplication and acts like a squared norm modulo torsion. Different projections of the points may have bounded heights or heights growing at very different rates.

The manuscript separates these scales by taking suitable quotients and normalizing each level independently. The subgroup relations then approach a limiting linear projector. Geometric positivity, built from Ax-type results and finitely many exceptional directions, constrains how this projector can annihilate the normalized points.

When every level diverges, a generalized Vojta–Rémond height inequality gives the contradiction. A bounded level needs a different argument: compare two limiting measures on a fixed low-height factor, one associated with a product map and another with consecutive differences. Their incompatible densities yield arithmetic positivity. Small sections then vanish to high order at the marked points, and a product theorem cuts out a smaller auxiliary variety, contradicting minimality.

The uniformity concerns the auxiliary varieties and their fields of definition; it does not assume that the marked intersection points have bounded degree.

Turn special structure into many controlled conjugates

M2 uses the gluing description of a mixed product surface. A product isogeny has degree $n^2$, while its rank-one degeneration length carries a reciprocal factor $1/n$. This produces a height estimate growing roughly linearly in $n$, up to the CM discriminant contribution. A balanced period construction on a related abelian threefold yields a stronger $n^{4/3}$ contribution. Comparing them forces the required Galois-orbit growth.

M3 and M4 instead study a rank-five quadratic system arising from trace-zero endomorphisms self-adjoint for the polarization. Extra endomorphisms supply positive two-dimensional planes. Period transport compares a fixed interior exceptional point with a varying target.

At suitable finite places, traces of endomorphism products give local equations. Comparing Frobenius-conjugate equations eliminates coefficients that depend on the prime. Grouping by conjugate data controls the number of equation types even when many primes contribute.

An interpolation and product-formula argument then gives a polynomial height bound in the field degree:

$$h(t)\le C\,[K(t):K]^\kappa.$$

Here $h$ is a fixed logarithmic height and $C,\kappa$ depend on the curve. Auxiliary level markings ensure that the equations are genuinely nonidentical; M4 uses a second marking to handle bad reduction at the chosen center.

Finally, endomorphism bounds convert height control into polynomially controlled arithmetic coordinates or large Galois orbits. Pila–Wilkie-type counting forces an algebraic family if there are too many such points. Monodromy or the established Daw–Orr implication then contradicts Hodge genericity. The elliptic-square argument is proved separately before being combined with the other branches.

7. Scope, limitations, and open questions

M1 concerns torsion cosets in abelian varieties over algebraic numbers. It does not by itself prove Zilber–Pink for every semiabelian variety, every transcendental base field or every mixed Shimura variety. Its finite-rank consequence retains the stronger no-proper-translate assumption.

The $\mathcal A_2$ conclusion is a theorem about curves. It does not cover arbitrary higher-dimensional subvarieties of every $\mathcal A_g$. M3 requires equality of the full endomorphism algebra with a quaternion division algebra; merely embedding such an algebra into something larger is not its stated condition.

No uniform or effective bound on the number of exceptional subvarieties is claimed. Finiteness alone does not provide a terminating procedure listing every exceptional point. Effective counts, explicit enumeration and extensions to broader moduli spaces remain additional problems.

8. Sources and further reading

M1’s Theorems 1.1–1.2 and Corollary 1.3 state the abelian and finite-rank conclusions. M2’s Theorems 1.1–1.2 give the mixed-product finiteness and Galois-orbit bound. M3’s Theorem 1.1 specifies the full quaternionic endomorphism condition. M4’s Theorems 1.1–1.3 separate the elliptic-square branch from the combined curve theorem.

The Habegger–Pila and Daw–Orr papers linked in section 3 give primary accounts of the earlier partial results and counting framework. The introductions of all four manuscripts carefully identify the height, period and monodromy inputs. Optional background on abelian varieties and complex multiplication can supplement the definitions above.

Source manuscripts

• M1. OpenAI, The abelian Zilber-Pink conjecture (September 24, 2026).
• M2. OpenAI, The E×CM component of Zilber-Pink for curves in A2 (September 24, 2026).
• M3. OpenAI, Quaternionic division points on curves in the Siegel threefold (September 24, 2026).
• M4. OpenAI, Elliptic squares and Zilber-Pink for curves in A2 (September 24, 2026).

Sources consulted on 10 October 2026 at repository revision fd4aeeb. Download the manuscript citations.