gpt 6 astra ultra attempt on Math-350

Mazur's finite-components conjecture for rational points

Definitions and mathematical statement

Let $V$ be an algebraic variety over ${\mathbb{Q}}$. Its real points $V({\mathbb{R}})$ carry the usual analytic topology, obtained in affine charts from Euclidean topology and glued across charts. Rational points embed naturally into real points. Mazur's conjecture in this form asserts \[ \#\pi_0\!\left(\overline{V({\mathbb{Q}})}^{\,V({\mathbb{R}})}\right)<\infty, \] where $\pi_0$ denotes connected components and the bar is closure in that real topology. The closure is not Zariski closure, and the rational points themselves need not be connected or finite. The assertion ranges over all varieties in the catalogue's algebraic convention; it imposes no effectiveness or uniform bound on the number of components as $V$ varies.

Formulation and concept references: [S1].

Short English statement

For every variety over the rationals, does the real-topological closure of its rational points have only finitely many connected components?

Research attempt: real closures, curves, and the projection obstruction

1. Target and source scope.

The object whose connected components must be counted is $\overline{V(\mathbb Q)}\subset V(\mathbb R)$, not $V(\mathbb R)$ itself and not the Zariski closure. The closure is taken in the usual real topology. Cornelissen and Zahidi's Conjecture 2.1 and Example 2.2, read in the working arXiv HTML version [E1] on 27 September 2026, state this exact conjecture, explain the curve case, and discuss consequences for Diophantine definitions. Their paper is historical evidence, not a current proof in arbitrary dimension. The present argument reproves the elementary topological steps and explicitly identifies its external inputs: Faltings's finiteness theorem for curves of genus at least two, basic real algebraic geometry, and the closed-subgroup theorem for Lie groups.

2. Smooth rational varieties provide a complete subcase.

Suppose $V$ is a smooth geometrically integral variety over $\mathbb Q$ and contains a dense Zariski-open set $U$ isomorphic over $\mathbb Q$ to a Zariski-open set $W\subset\mathbb A^d$. Rational points are dense in $W(\mathbb R)$: every sufficiently small Euclidean ball about a point of $W(\mathbb R)$ stays in $W(\mathbb R)$, and $\mathbb Q^d$ meets that ball. The isomorphism transports this density to $U(\mathbb R)$.

We also need that $U(\mathbb R)$ is dense in $V(\mathbb R)$. At a real smooth point, real analytic coordinates identify a neighborhood with an open subset of $\mathbb R^d$. A proper algebraic closed subset cannot contain an open subset of that neighborhood. Indeed, a regular function vanishing on such a real open set has identically zero analytic germ; by complexification and the identity theorem it vanishes on a complex open set, hence on the irreducible variety. This contradicts properness. Applying this to local equations for $V\setminus U$ shows that the latter has empty real interior. Therefore \[ \overline{V(\mathbb Q)}=V(\mathbb R). \] The real points of a finite-type real variety have finitely many connected components, by semialgebraic finiteness. This proves the target for smooth rational $V$. If $V(\mathbb R)$ is empty the conclusion is immediate. In dimension zero the assertion is also immediate.

Smoothness is doing work in the density step. A singular real locus may contain isolated real points lying outside a chosen birational chart. A proof for smooth rational varieties must not claim, without a separate argument, that every real point of every rational singular variety is approximated by the chart.

3. A complete curve reduction, including missing points and singularities.

First consider a smooth projective geometrically integral curve $C$. If its genus is at least two, Faltings's theorem says $C(\mathbb Q)$ is finite, so its real closure is finite. This deep input is used as a theorem, not reproved by a height computation. If the genus is zero and there is a rational point, $C\cong\mathbb P^1$ over $\mathbb Q$, and the preceding density argument applies. If there is no rational point, the relevant closure is empty.

In genus one, a rational point identifies $C$ with an elliptic curve $E$ over $\mathbb Q$. The closure $H=\overline{E(\mathbb Q)}$ is a closed subgroup of the compact real Lie group $E(\mathbb R)$. The identity component of $E(\mathbb R)$ is a circle, and $E(\mathbb R)$ has at most two components. Here is the needed elementary circle fact. If $H_0$ is a closed subgroup of $\mathbb R/\mathbb Z$, its inverse image $G\subset\mathbb R$ is a closed additive subgroup containing $\mathbb Z$. If $G$ has positive elements tending to zero, integer multiples approximate every real number and closedness gives $G=\mathbb R$. Otherwise its least positive element exists, generates $G$, and divides one. Thus $H_0$ is either the whole circle or finite. Applying this to $H\cap E(\mathbb R)^0$, and then to its finitely many cosets, proves that $H$ has finitely many components.

For a smooth nonprojective curve, take its smooth projective completion and remove a finite set of geometric points. On real points this removes finitely many points. A finite subset stays finite after removal. A finite union of circles and points acquires only finitely many components after removal of finitely many points. Moreover the closure inside the open curve is the intersection of the projective closure with that open curve: deleting finitely many rational points from a dense infinite subgroup does not remove any of its accumulation points, and the finite case is direct. Hence the assertion remains true.

Finally, for a reduced geometrically integral singular curve, normalization is an isomorphism outside finitely many singular points. The preceding result applies on this common smooth open set, and adding the finitely many rational singular points cannot create infinitely many connected components. Curves with several geometric components require one further descent observation: a rational point on a component not individually defined over $\mathbb Q$ lies on its intersections with its Galois conjugates, except when the component is Galois invariant. Those intersections are finite for distinct curve components. Treat the finitely many components defined over $\mathbb Q$ and add the remaining finite set. Nilpotent structure does not change the real or rational point sets. This completes the curve subcase with the finite-type qualifications explicit.

4. Proper abelian varieties and products.

For an abelian variety $A/\mathbb Q$, the set $A(\mathbb R)$ is a compact Lie group: projectivity gives compactness and smoothness gives its manifold structure. The closure $H=\overline{A(\mathbb Q)}$ is a closed subgroup, since multiplication and inversion are continuous. The closed-subgroup theorem makes $H$ a Lie subgroup. Its identity component is open in $H$, so the quotient by that component is discrete. A compact discrete space is finite. Consequently $H$ has finitely many connected components. Notice that this proof does not need a choice of Mordell--Weil generators or an effective density computation.

If the conjecture is known for $V$ and $W$, it holds for their product. In fact \[ (V\times W)(\mathbb Q)=V(\mathbb Q)\times W(\mathbb Q), \qquad \overline{V(\mathbb Q)\times W(\mathbb Q)} =\overline{V(\mathbb Q)}\times\overline{W(\mathbb Q)}. \] The latter equality follows from the product basis for the topology. Products of connected sets are connected, and projecting a connected set to either factor preserves connectedness. Therefore components of the product are exactly products of components. Their number is the product of two finite numbers. These arguments supply genuine higher-dimensional examples, but not a reduction of arbitrary varieties to abelian or rational ones.

5. A conditional obstruction to discrete Diophantine subsets.

Assume the full conjecture. Let $V/\mathbb Q$ be affine and let $\pi:V\to\mathbb A^1$ be a morphism over $\mathbb Q$. Put $S=\pi(V(\mathbb Q))$, and suppose the real closure $\overline S$ is totally disconnected. We claim that $S$ must be finite.

Let $H=\overline{V(\mathbb Q)}$. Continuity implies \[ \pi(H)\subseteq\overline{\pi(V(\mathbb Q))}=\overline S. \] Every connected component of $H$ has connected image, hence has a singleton image because $\overline S$ is totally disconnected. By the conjecture there are only finitely many such components, so $\pi(H)$ is finite. Since $S\subseteq\pi(H)$, the claim follows. In particular, an existential polynomial definition of $\mathbb Z$ inside $\mathbb Q$ would violate the conjecture, because $\mathbb Z$ is an infinite closed discrete subset of $\mathbb R$. Several polynomial equations can be encoded as an affine variety, and existentially quantified variables are exactly the coordinates forgotten by $\pi$.

This proof uses only the inclusion displayed above. It does not assert that projection commutes with closure. Such an equality is false even for simple real algebraic sets: the projection of $xy=1$ to the $x$-axis omits zero, while its image closure contains zero. The corresponding points escape to infinity in the $y$ coordinate. Properness would prevent that particular escape, but properness has not been assumed in the Diophantine application.

6. Two attractive reductions that do not work.

First, real algebraic sets have finitely many components, but their closed subsets need not. For example, $\{0\}\cup\{1/n:n\geq1\}$ is a closed subset of the connected real affine line and has infinitely many components. The closure of rational points is a special closed subset; semialgebraic finiteness applies to it only after one proves that it has suitable definability properties. No such general definability theorem has been obtained here.

Second, a dominant map from a variety already understood does not ensure that all rational points of the target lift to rational points. Even $x\mapsto x^2$ on the affine line misses many positive rational numbers. A birational isomorphism on a specified open set does preserve rational points there, which is why Section 2 works. Replacing that isomorphism by a general dominant map loses the arithmetic information. Likewise a finite cover may have no rational point above most rational points of its base.

7. Diagnostics and remaining gap.

The saved exact diagnostic illustrates the projection issue using the rational points $(1/n,n)$ of $xy=1$, and verifies finite products of explicitly represented finite component sets. It makes no claim to count components of a general arithmetic closure. The curve and group arguments are deductive, and their deep imported inputs are identified above.

UNRESOLVED. The target is proved for curves, smooth rational varieties, proper abelian varieties, and finite products of these examples. The conditional Diophantine obstruction is established without an invalid closure-projection equality. The first general gap is to control the topology of the arithmetic closure for varieties with neither a dense rational chart nor a compatible compact group structure; finiteness of the ambient real topology alone gives no such control.

Sources

• S1 Gunther Cornelissen and Karim Zahidi, Topology of Diophantine Sets: Remarks on Mazur's Conjectures $2000$. https://ar5iv.labs.arxiv.org/html/math/0006140. Formulation source.
• E1 G. Cornelissen and K. Zahidi, Topology of Diophantine Sets: Remarks on Mazur's Conjectures, working arXiv HTML, Conjecture 2.1, Example 2.2 and Section 3. https://arxiv.org/html/math/0006140. Read 27 September 2026; alternate accessible rendering of S1. Historical formulation and curve/Diophantine discussion, not a general resolution.