Shafarevich refuted?

0:00Covers and convexity

Every compact Riemann surface is covered by one of three simple spaces: the sphere, the plane, or the disk. Shafarevich asked, in his textbook on algebraic geometry, whether something similar holds in every dimension: is the universal cover of every smooth projective variety holomorphically convex? Holomorphically convex means this. Take any compact set K. Collect the points where every holomorphic function is no bigger, in absolute value, than its largest absolute value on K. That hull must be compact again. Then the cover maps properly onto a Stein space, a space with plenty of holomorphic functions, and the fibers are compact. So if the cover contains no compact curves or anything larger, the conjecture says it should itself be Stein. The evidence was strong. Katzarkov proved it for virtually nilpotent fundamental groups, and Eyssidieux, Katzarkov, Pantev and Ramachandran proved it whenever the fundamental group has a faithful linear representation. Bogomolov and Katzarkov proposed possible counterexamples, but they depended on group theory that was never proved.

1:10Two obstructions

Here is the obstruction. Take a connected cluster of rational curves, spheres meeting at nodes, where the meetings form closed loops. Each sphere is simply connected, but those loops can have infinite image in the fundamental group. Then in the universal cover, each sphere lifts to a compact sphere, but the cluster unrolls into an infinite necklace. A holomorphic function is constant on each compact sphere, and neighbors share a point, so it takes one value on the whole necklace. The hull of a single point then contains the entire necklace, which is not compact. The cover is not holomorphically convex. The second obstruction is topological. Put an abelian surface, a torus of real dimension four, inside the variety, so that its fundamental group, Z⁴, has image Z in the variety’s fundamental group. In the universal cover it becomes a line times a three-torus, which has nonzero homology in degree three. Inside a Stein manifold it would be a Stein surface, with the homotopy type of a two-dimensional complex, so the cover cannot be Stein.

2:16The claim

Result 046 consists of two OpenAI preprints. The first, 26 pages dated September 23, claims a smooth projective surface containing a connected cluster of rational curves with infinite image in its fundamental group. If it holds, Shafarevich’s conjecture fails in dimension two. The second, 41 pages dated October 5, claims a smooth projective fourfold with large fundamental group, so its universal cover contains no compact curves or anything larger, while a simple abelian surface inside it has infinite cyclic image. The cover is then not Stein, which would refute even the large fundamental group form of the conjecture.

3:01How the constructions go

The surface starts from the elliptic curve with a symmetry of order three. Multiplication by four and the sign quotient give a family of spheres with 48 moving punctures. At one parameter value they collide in 24 pairs, and a base change and blowups turn that fiber into a main sphere with 24 spheres attached. The hard part is a quotient group that stays infinite after every relation is imposed. The tool is the Golod–Shafarevich inequality, from the same Shafarevich: a pro-2 group is infinite if, at some weight t, its weighted generators exceed its weighted relations by more than one. Forty-eight generators, 24 relations for the sign symmetry, and at t = 49/200 that leaves a margin of almost three hundredths. One hundredth goes to deep monodromy relations, and one hundredth to high powers of the boundary loops. Then a finite cover, chosen so that no local boundary group loses information, gives a smooth projective surface. There the special fiber becomes a cluster of spheres whose meetings form loops, and those loops carry the infinite image. The fourfold starts from a genus-two curve with a simple Jacobian. Arrangements of hyperplanes in complex hyperbolic ball quotients impose relations that make the curve’s image cyclic. Local covers modeled on the infinite dihedral group, read along hyperbolic graphs, are used to argue that this image is infinite. Two copies of the curve, folded by an order-four symmetry, lead to its Jacobian, a simple abelian surface, whose image the paper argues is infinite cyclic too. Cutting its ambient variety, times a six-dimensional abelian variety, down to dimension four gives the fourfold.

4:53Where it stands

These are two preprints, not yet peer reviewed, with no Lean proof, and neither has been revised. The fourfold paper cites nothing else from the release; the surface paper cites two release papers, only for side remarks. We found no public assessment. If they hold, the universal cover of a smooth projective variety need not be holomorphically convex, already for a surface, and not even when it contains no compact curves or anything larger.