The international mathematics community has long regarded the 3D Ghys conjecture as a formidable wall, unsolved for over thirty years. Now, a Chinese research team has not only climbed that wall but did so using a domestically developed large AI model. This marks a significant pivot: for the first time, Chinese scientists have relied on a homegrown LLM to crack a problem of this magnitude, challenging the assumption that frontier mathematical reasoning requires Western proprietary models.

The Intern-S2 Collaboration

The breakthrough was achieved by a team led by Professor Xu Disheng of the Greater Bay University, who also serves as a visiting scholar at the Shanghai Artificial Intelligence Laboratory. They utilized the lab's Intern-S2 science large model to handle the heavy lifting of large-scale, long-chain, and highly complex mathematical derivations. The human researchers were not passive; they steered the overall direction and rigorously screened lines of reasoning, creating a hybrid workflow where the AI generated the depth and the humans provided the strategic oversight. After multiple rounds of iterative interaction, this partnership produced a complete and rigorous proof.

Rigorous Machine Verification

Skepticism is the default stance for any AI-generated mathematical claim, so the team brought in independent verification from Peking University. Led by Liu Ruochuan, an academician of the Chinese Academy of Sciences, and Professor Dong Bin, this verification team employed a self-developed framework and massive parallel computing with AI agents. They generated hundreds of thousands of lines of verification code to formally check every core theorem. The result was a rigorous confirmation that the entire proof is logically complete, gap-free, and fully valid, moving beyond mere plausibility to machine-verified certainty.

Implications for Advanced Reasoning

First proposed in 1995 by French mathematician Étienne Ghys, the conjecture sits at the heart of complex dynamics and Anosov systems, as illustrated by chaotic examples like the Arnold cat map. The source notes that its real-number counterpart remains an open problem worldwide, serving as a key benchmark for measuring advanced mathematical reasoning ability. By solving the 3D version, the team has demonstrated that the Intern-S2 model possesses reasoning capabilities that rival or exceed current state-of-the-art expectations for formal mathematics. The paper has been released on an international preprint platform, and crucially, the verification code has been fully open-sourced, allowing the global community to audit the work.

Key Takeaways

  • The 3D Ghys conjecture, unsolved for 30+ years, was proved using the domestic Intern-S2 model.
  • Verification involved hundreds of thousands of lines of code from a Peking University team.
  • The real-number counterpart of the conjecture remains an open global benchmark.
  • All verification code is open-sourced, setting a new standard for transparency in AI math proofs.

The Bottom Line

This isn't just a math win; it's a geopolitical signal that domestic LLMs are no longer playing catch-up in formal reasoning, forcing Western labs to rethink their benchmarks.