Family 040 — Bloch’s conjecture: when points on a surface have no hidden differences

← All digests

Family 040 · Algebraic and complex geometry
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-12

1. The question

Take two points on a smooth complex algebraic surface. Topologically, they represent the same class: a point is a point. Algebraic geometry asks a finer question. Can one transform one into the other using the zeros and poles of rational functions on curves lying in the surface?

On the projective plane the answer is yes. Join the points by a line, identify that line with the projective line, and choose a rational function with a zero at one point and a pole at the other. On more complicated surfaces, such transformations can leave a large collection of differences that topology cannot see.

Bloch’s conjecture predicts when these hidden differences disappear, or are completely measured by an associated complex torus. Family 040 contains one manuscript claiming the conjecture for every smooth projective complex surface with no holomorphic two-forms. Its central new case has no holomorphic one-forms either. In that case, it claims that the only invariant of a finite integral combination of points is its total coefficient.

The strength lies in the word algebraic: an equality of topological classes would be automatic and would not answer the question.

2. Background and essential definitions

Points, functions, and the Chow group

A zero-cycle on a surface $S$ is a finite formal sum

$$z=\sum_i n_i[p_i],\qquad n_i\in\mathbb Z.$$

Its degree is $\deg z=\sum_i n_i$. If a rational function on a curve has zeros at some points and poles at others, with their multiplicities, its divisor defines a degree-zero cycle. Rational equivalence allows us to add or subtract such divisors, including their images from curves mapping into the surface.

The resulting group is the Chow group $\operatorname{CH}_0(S)$. Its subgroup of degree-zero classes is $\operatorname{CH}_0(S)^0$. Thus $[p]-[q]=0$ in this group means an actual finite chain of algebraic relations exists, although the relation need not come from a single curve through the two points.

Saying $\operatorname{CH}_0(S)=\mathbb Z$ via the degree map means precisely that every degree-zero class vanishes. It does not mean the surface is the projective plane, or even that the surface can be parametrized by rational functions.

Two numerical measurements of differential forms

The geometric genus and irregularity are

$$p_g(S)=\dim_{\mathbb C}H^0(S,\omega_S), \qquad q(S)=\dim_{\mathbb C}H^1(S,\mathcal O_S).$$

Here $\omega_S$ is the line bundle of holomorphic two-forms; $\mathcal O_S$ is the sheaf of holomorphic functions. For a smooth projective complex surface, $q$ also counts independent holomorphic one-forms. These are different invariants: assuming $p_g=0$ does not automatically mean $q=0$.

The Albanese variety $\operatorname{Alb}(S)$ is a complex torus, in fact an abelian variety, of dimension $q$. Integrating one-forms along paths produces the Albanese map on degree-zero cycles,

$$\operatorname{alb}_S:\operatorname{CH}_0(S)^0 \longrightarrow \operatorname{Alb}(S)(\mathbb C).$$

Bloch’s prediction is that when $p_g=0$, this map has no kernel. It is already surjective. When also $q=0$, its target is zero, so the prediction reduces to $\operatorname{CH}_0(S)^0=0$.

3. Development history and previous partial results

Mumford’s work on rational equivalence of zero-cycles showed that a nonzero holomorphic two-form forces the group of zero-cycles to be too large to be described by a finite-dimensional parameter space in the relevant sense. Differential forms therefore impose a real obstruction to simple algebraic relations between points. Bloch’s conjecture asks for a converse after accounting for the Albanese variety.

An early broad theorem of Bloch, Kas, and Lieberman established the prediction for surfaces with $p_g=0$ that are not of general type. A surface is of general type when its pluricanonical sections grow as fast as the dimension permits. These surfaces form the part of the classification where the canonical bundle carries the richest geometry. They remained the main difficulty.

Many general-type examples were then settled using their special constructions or symmetries. Work of Inose–Mizukami, Barlow, and later authors exploited group actions, coverings, and correspondences between surfaces. Numerical Godeaux, Campedelli, Burniat, and Inoue surfaces supplied important testing grounds, but a collection of named constructions does not exhaust all surfaces with the same numerical invariants.

Voisin’s work on Catanese and Barlow surfaces developed another route through families and correspondences. It showed that one could obtain zero-cycle relations without requiring each individual surface to display an especially simple symmetry. This was a major enlargement of the methods available, while still relying on geometric properties of the families involved.

The present manuscript also discusses overlapping recent proposed solutions, including Guletskii’s 2025 preprint and a 2026 proposal of Banerjee for the general-type case with canonical self-intersection nine. It presents its own argument through moduli spaces of quotient sheaves. The account here follows those attributions; it does not settle priority or the correctness of competing proofs.

The remaining conceptual gap was a uniform mechanism that works without a special presentation of the surface. The manuscript proposes such a mechanism by constructing an algebraic relation for its diagonal.

4. What the new manuscripts claim

An integral theorem for every regular surface in scope

Theorem 1.1 of M1 states: if $S$ is a smooth connected projective surface over $\mathbb C$, and

$$p_g(S)=q(S)=0,$$

then the degree homomorphism is an isomorphism

$$\deg:\operatorname{CH}_0(S)\xrightarrow{\ \sim\ }\mathbb Z.$$

The coefficients here are integers. There is no assumption that $S$ is minimal, simply connected, or obtained from a particular covering construction. Every pair of complex points would have equal classes, and every degree-zero integral cycle would vanish.

Corollary 1.2 combines this with the earlier non-general-type theory to give the full stated surface formulation: whenever $p_g(S)=0$, the Albanese map is an isomorphism. The possibility $q>0$ is retained in this formulation; one must not replace its conclusion by $\operatorname{CH}_0(S)=\mathbb Z$.

Consequences for cycles on larger spaces

For $p_g=q=0$, M1 also states that all rational algebraic cycles on products of $S$, and on its Hilbert schemes of points, are determined by their cohomology classes. A Hilbert scheme $S^{[n]}$ parametrizes length-$n$ finite subschemes, including configurations where points merge with infinitesimal structure.

More precisely, for $Y=S^m$ or $S^{[n]}$, the cycle-class maps are isomorphisms

$$\operatorname{CH}^i(Y)\otimes\mathbb Q \xrightarrow{\ \sim\ }H^{2i}(Y,\mathbb Q), \qquad H^{2i+1}(Y,\mathbb Q)=0.$$

The manuscript derives these statements from established results on Chow motives, which package algebraic correspondences into a common structure. It identifies the rational motive of $S$ as a sum of elementary pieces. These further consequences have rational coefficients; the integral claim above concerns zero-cycles on $S$ itself.

5. Why the advance matters

Importance: 5/5. If correct, this would settle a central conjecture about the gap between algebraic cycles and their cohomological shadows, across an entire class of surfaces.

Even on a surface with $p_g=q=0$, having only algebraic cohomology classes does not immediately determine rational equivalence. Homologically invisible cycles are precisely what makes the subject difficult. The claimed theorem would show that in this setting there is no residual degree-zero information at all.

The generality also matters. Numerical invariants can be shared by surfaces with very different fundamental groups and constructions. A theorem applying uniformly would replace many separate calculations with one principle. Its consequences for all powers and Hilbert schemes would propagate that principle into arbitrarily high dimensions, though only for these particular spaces.

The proposed method is significant independently of the list of examples: it uses enumerative geometry to create a relation in the Chow group itself, retaining information that an ordinary numerical integral would discard.

6. The main idea

Make the identity correspondence vanish on degree-zero cycles

The diagonal $\Delta_S\subset S\times S$ consists of pairs $(x,x)$. Viewed as a correspondence, it acts as the identity on cycles. M1 seeks a relation of the form

$$0=c[\Delta_S]+\Gamma \quad\text{in }\operatorname{CH}^2(S\times S)\otimes\mathbb Q, \qquad c\ne0,$$

where $\Gamma$ is a sum of products of cycles from the two separate copies of $S$. Those product terms act trivially on degree-zero zero-cycles. Applying the relation would therefore give $c z=0$ for every such rational cycle $z$, hence its vanishing.

This must be a Chow relation, not merely a cohomology relation. The latter would erase the information one is trying to control.

Keep two marked outputs in a moduli-space calculation

The construction uses a Quot scheme, a parameter space for quotient sheaves of a fixed bundle. Here the starting bundle is a sum of two line bundles, equipped with opposite weights for a one-dimensional torus action. Fixed points of that action split the quotient into simpler pieces involving divisors and finite subschemes.

Such parameter spaces can have singularities and unexpected dimension. A virtual class supplies the intersection-theoretic replacement for a fundamental class, using the specified deformation and obstruction data. Localization expresses its contribution through the torus-fixed pieces.

The crucial design choice is to retain two marked outputs on the surface. Instead of integrating everything to a number, the calculation produces a cycle on $S\times S$. Recursive formulas for Hilbert schemes then separate contributions where the outputs coincide, giving the diagonal, from contributions that factor through the two independent surfaces.

Show the diagonal really occurs

Obtaining a relation with coefficient $c=0$ would prove nothing. The manuscript reduces to minimal general-type surfaces and treats two numerical cases for $K_S^2$, the self-intersection of the canonical divisor.

For $K_S^2\le8$, it uses the intersection lattice and a limited number of point blowups to choose a class of negative square. Pairing the constructed relation with that class expresses the diagonal coefficient through nonnegative squares, with a nonempty positive contribution.

For $K_S^2=9$, that lattice direction is unavailable. M1 instead compares universal coefficients with an auxiliary Cartwright–Steger surface having the same relevant Chern numbers but nonzero holomorphic forms. Its geometry detects the needed coefficient. The two branches are parts of the same strategy, but neither is presented as a substitute for the other.

Finally, the Chow relation initially gives vanishing with rational coefficients. An integral class could still be torsion, meaning some nonzero integer multiple vanishes. Roitman’s torsion theorem, together with $q=0$, removes that remaining possibility. Birational invariance returns the conclusion from the modified surface to the original one.

7. Scope, limitations, and open questions

The theorem concerns smooth projective surfaces over the complex numbers. It does not assert the analogous statement for arbitrary singular surfaces, nonprojective compact complex surfaces, or varieties over fields of positive characteristic.

The assumption $p_g=0$ is structural: Mumford’s result explains why positive geometric genus leads to a different picture. When $q>0$, the Albanese variety remains as genuine information. The conclusion is then that it captures all degree-zero classes, not that those classes vanish.

The integral surface theorem and the rational motivic consequences should also remain separate. The latter do not claim integral cohomology determines every cycle on every product. Nor does the theorem supply an efficient algorithm that, given two points, prints a short collection of curves and rational functions relating them.

These are the manuscript’s stated results and approach. Questions beyond its scope include corresponding principles for higher-dimensional varieties, arithmetic fields, and more explicit descriptions of the relations between individual points.

8. Sources and further reading

M1 contains the integral statement in Theorem 1.1, the Albanese formulation in Corollary 1.2, and the rational motivic consequences in Corollary 1.3. Its introduction explains the earlier surface cases and the distinction between the two coefficient calculations.

The historical papers linked above provide the underlying questions and partial results. For a first encounter, the definitions of a Chow group and an Albanese variety are useful companions. The essential point to retain is that rational equivalence carries much more information than equality in cohomology.

Source manuscripts

• M1. OpenAI, Bloch’s conjecture for surfaces with p_g=0 (September 24, 2026).

Sources consulted on 12 October 2026 at repository revision fd4aeeb. Download the manuscript citations.