A preprint by Shanmu Jin, titled The Numerical Range Is a 2-Spectral Set, states that it proves Crouzeix’s conjecture, placing a striking claim in the public record. The manuscript was submitted August 6 and posted August 7, and its page explicitly notes that it is not peer-reviewed. The preprint identifies Jin as the author and is publicly available through the preprint server.
According to the South China Morning Post, Jin is a Peking Union Medical College Hospital postdoctoral researcher and a resident in Beijing, and he pursued the mathematical result while investigating questions related to transcranial ultrasound. The South China Morning Post also reports that a 16-hour autonomous run of GPT-5.6-Sol in ChatGPT Work assisted the work, framing a notable instance of AI woven into a research workflow. Readers can consult SCMP’s coverage for those details at the South China Morning Post.
ChatGPT-assisted preprint on Crouzeix’s conjecture
Jin’s manuscript arrives with a firm declaration in its title and body. The preprint says it proves Crouzeix’s conjecture, a problem of broad interest in numerical analysis circles, and presents the argument under the heading The Numerical Range Is a 2-Spectral Set. The document’s listing shows it was submitted August 6 and posted August 7, and it makes plain that the manuscript has not undergone peer review. That status matters for how the community will process the claim, since the preprint server page itself highlights the absence of formal evaluation.
The South China Morning Post provides the initial biographical anchor and the narrative of how the work unfolded, identifying Jin as a postdoctoral researcher at Peking Union Medical College Hospital and a resident in Beijing. SCMP reports that Jin was exploring issues tied to transcranial ultrasound when he worked on the mathematical question, and it attributes a portion of the effort to a 16-hour autonomous run of GPT-5.6-Sol in ChatGPT Work. These details, as reported by SCMP, situate the paper at the junction of medicine, mathematics, and AI-enabled tooling, while keeping the source of those specifics clear and attributed.
What outside reports say about expert reactions
Interesting Engineering reports that mathematicians Alex Townsend, Anne Greenbaum, and Michel Crouzeix reviewed the manuscript and considered the proof correct. That account, while notable, remains third-party reporting and is not a formal peer-review outcome. It should be read alongside the explicit label on the preprint that it is not peer-reviewed. Readers can find that coverage at Interesting Engineering.
The status of Jin’s paper therefore sits in a familiar phase for high-profile mathematical claims. It is publicly posted, it asserts a result of interest, and it has generated reported responses. At the same time, the manuscript has not completed peer review. The combination of a visible preprint, reported expert commentary, and a clearly stated lack of peer review invites further scrutiny and discussion that will unfold through established channels.
How attribution and process frame the story
The preprint’s own framing and the reporting around it offer two anchors for readers. First, the manuscript states the claim directly and provides the argument under Jin’s name, which allows mathematicians to examine the work on its merits. Second, SCMP’s account supplies context about the research setting and the AI assistance, specifically that a 16-hour autonomous run of GPT-5.6-Sol in ChatGPT Work assisted the work. Each of these components is part of the present record, and each is properly attributed to its source.
In parallel, Interesting Engineering reports that three mathematicians, including Michel Crouzeix, reviewed the manuscript and considered the proof correct. That is a specific and newsworthy report, yet it is distinct from a journal’s peer-review process. The preprint server’s label that the manuscript is not peer-reviewed remains the operative status note. It is important to hold those pieces together without overshooting what has been independently verified. The article does not assert that AI independently solved the conjecture, and it does not state that the result is finally settled by publication. It surfaces that the author’s account and outside reporting describe how ChatGPT factored into the workflow, and it underscores the current stage of review.
For readers tracking how AI tools are entering research practice in China and beyond, EastFrontier has covered adjacent developments in the country’s AI landscape. Related EastFrontier coverage can be found here and here. These links provide additional reporting from our newsroom that may help situate the present story within the broader flow of AI-related news.
Ultimately, the elements that can be stated with confidence rest on the sources at hand. The preprint says it proves Crouzeix’s conjecture and confirms that the work is not peer-reviewed. SCMP identifies Jin’s institutional role and location and reports that he worked on the result while examining questions related to transcranial ultrasound, with a 16-hour autonomous run of GPT-5.6-Sol in ChatGPT Work assisting the process. Interesting Engineering reports that several mathematicians reviewed the manuscript and considered the proof correct, and it emphasizes that this is not a formal peer-review outcome. These points describe the state of play as it currently stands.
As the community reads and tests the argument, attention will likely remain on the clarity of attribution and the path to validation. The combination of a public preprint, reported use of a generative AI tool within ChatGPT Work, and coverage that brings mathematicians’ reported reactions into view ensures that the discussion is grounded in citable sources. Whatever unfolds next, the present documentation offers a clear starting point for following the claim, its assessment, and its reception.