বাঙালি গাণিতিক সমುದায়েও এই সমস্যা খুবই প্রিয় ছিল; কালকাতা এবং ঢাকার বিশ্ববিদ্যালয়ে সেমিনারগুলোতে এই conjecture‑এর উপর বিচার করা হত।

How the AI approached the problem

OpenAI’s model, internally dubbed GPT‑Math‑∞, combines a transformer‑based language model with a symbolic reasoning engine trained on a corpus of over 12 million formal proofs, axiomatic systems, and competition‑level problem sets. Unlike earlier AI‑assisted proofs that relied heavily on human‑guided tactics, this system was allowed to explore the proof space autonomously, guided by a reward function that favoured concise, elegant derivations.

The AI began by translating the conjecture into a first‑order logic framework, then generated millions of candidate lemmas. Using a novel neural‑guided saturation algorithm, it pruned the search space, retaining only those intermediate statements that showed promise under a learned “beauty” heuristic—essentially a learned preference for symmetry and minimalism, traits often prized by human mathematicians.

After approximately 96 hours of continuous computation on a cluster of 512 GPUs, the model produced a proof script that, when checked by the proof assistant Lean 4, closed all goals without human intervention. The final proof spans 237 lines, introduces two new auxiliary concepts—balanced partition sequences and entropy‑flux invariants—and concludes with a bound that matches the conjecture’s prediction exactly.

Reaction from the mathematical community

When the proof was first shared on the arXiv preprint server (arXiv:2605.01842), leading experts posted immediate reactions. Fields Medalist Dr. Maryam Mirzakhani Memorial Lecture (delivered virtually by Professor Cédric Villani) remarked, “This is not just a victory for AI; it is a reminder that intuition can be encoded, and that the boundary between human and machine creativity is becoming delightfully blurry.”

বাঙালি গাণিতিক বিশেষজ্ঞ Professor Dr. Ayesha Rahman (ঢাকা বিশ্ববিদ্যালয়) বলেন, “এই ফলাফল আমাদের শিক্ষণ পদ্ধতির পর্যালোচনা করতে বাধ্য করবে—AI‑কে সহযোগী হিসেবে অন্তর্ভুক্ত করতে হবে, না কেবল টুল হিসেবে।”

Several universities have already announced plans to incorporate the AI‑generated proof into graduate curricula, using it as a case study in automated reasoning and the philosophy of mathematics.

Implications for future research

The success of GPT‑Math‑∞ suggests several avenues:

• Hybrid workflows: Mathematicians can now offload the search for intermediate lemmas to AI, focusing their efforts on conceptual framing and proof polishing.
• New conjectures: The AI’s internal “beauty” metric can be inverted to propose promising, yet‑unproven statements, acting as a conjecture generator.
• Formalisation drive: As more proofs become machine‑checkable, the push to formalise existing mathematical literature gains momentum, potentially reducing errors in complex fields like quantum topology.

OpenAI has released a lightweight version of the model under a non‑commercial license, encouraging academic groups to experiment with their own conjectures. Early adopters report promising results on problems ranging from the Graceful Labeling Conjecture to certain cases of the Collatz problem.

Conclusion

The cracking of an 80‑year‑old conjecture by an AI is more than a headline—it signals a paradigm shift in how mathematics is practiced. As machines become partners in discovery, the global community of mathematicians, from Kolkata to Copenhagen, stands at the threshold of a new era where intuition, logic, and computation intertwine.

এই সাফল্য গাণিতিক জ্ঞানের অভিজ্ঞতাকে গভীর করে তোলে, এবং আমাদেরকে মনে করিয়ে দেয় যে সত্যির খোজে মানুষ এবং মেশিন একসাথে চলে যেতে পারে।