A research team led by Rahul Saha published a case study on arXiv on 11 August 2026 documenting collaboration between humans and an AI research system in mathematical research. The work records improvements to the known bounds of the Grothendieck constant KG – a problem that has been open for approximately 70 years. The team narrowed the lower bound to 6π/11 and the upper bound to π²log(1+√2) − 10⁻⁴.
The findings made it possible to determine the first decimal place of KG for the first time. This is 7, as detailed in the accompanying mathematical publication "New Lower and Upper Bounds for the Grothendieck Constant" (arXiv:2608.11158, published 12 August 2026). The case study was authored by Alan Li, Rahul Saha, Anton Xue, Swarat Chaudhuri, Adam Klivans, Pravesh K Kothari and Raghu Meka.
What is the Grothendieck Constant?
The Grothendieck constant KG quantifies the threshold between difficult combinatorial optimization problems and their efficient continuous relaxations. The precise value of the constant is unknown. It plays a fundamental role in complexity theory and was first described in the 1950s by Alexander Grothendieck.
Methodological Approach: AI System Under Human Direction
The team developed an AI research system that was asynchronously controlled by human operators and specifically tailored to the Grothendieck constant problem. Domain experts recognized the AI's insights as original, according to the authors. The case study discusses strengths and weaknesses of AI use in mathematical research, ideal conditions for AI breakthroughs, and the practical research process over longer time horizons.
On 14 August 2026, Rahul Saha announced the breakthrough on X. He wrote: "Academics should shape AI-assisted research to keep it fundamentally human." The team deliberately published both components – the mathematical results and the AI methodology – to enable the academic community to shape AI-assisted research.
Publication Details and Classification
The case study (arXiv:2608.11195) appeared on 11 August 2026 at 17:53:48 UTC. A revised version (v2) followed on 13 August 2026 at 17:46:18 UTC. The paper is classified in the categories Computational Complexity (cs.CC), Artificial Intelligence (cs.AI) and Mathematics (math).
Context: AI Use in Mathematical Research
The work addresses two central questions: how AI agents can be effectively deployed in mathematical research and what progress is possible on classical open problems. The improvement of the Grothendieck constant bounds represents progress on a problem that has been open for over seven decades. The study documents collaboration between humans and AI over an extended research period, without the authors providing specific time details.
The mathematical results are described in the separate publication "New Lower and Upper Bounds for the Grothendieck Constant", which became available on arXiv on 12 August 2026. Both publications are freely accessible.