A Chinese artificial intelligence model helps prove a conjecture open for thirty years

A Chinese research group led by Professor Xude Sheng has announced it has solved the three-dimensional Ghys conjecture, a problem open for thirty years, using the domestic artificial intelligence model Shuying-S2. This is reported by Pandaily and KuCoin News, both based on the official statement from the Shanghai Artificial Intelligence Laboratory: the news therefore currently stems from a single primary source, the laboratory itself; no independent confirmation is available at this time. This is a story that this morning’s edition had not yet reported.

A team from Peking University validated the proof using the SHEAF framework. The study, available on arXiv and on GitHub, is presented as the first time an artificial intelligence model developed in China has contributed to a mathematical discovery of global relevance.

The conjecture, proposed by French mathematician Étienne Ghys in 1995, concerns holomorphic Anosov systems: it posits that every system of this type defined on a compact complex manifold is biholomorphically conjugate to a holomorphic affine automorphism on a particular complex infra-nilmanifold. It is a statement in complex geometry that remained unproven for three decades.

According to the reconstruction provided by the sources, starting in August the model began exploring various approaches to proving the conjecture, producing thousands of pages of mathematical exposition before arriving at the formulation later validated by the Peking University team.

The case fits into a line of research exploring the use of artificial intelligence models as tools supporting mathematical proof, working alongside human researchers rather than replacing them: the process described in fact involves collaboration between the model and the mathematicians, with a final phase of formal validation conducted by a university team independent of the laboratory that developed the model.

The case comes as several laboratories around the world step up efforts to show that artificial intelligence models can contribute to original scientific discoveries, not merely automate already-known tasks. It remains to be seen whether the international mathematical community, outside the groups involved in the validation, will independently confirm the proof once it is published in full.