AI agent rediscovers geometric invariant with high precision fitting
Can an AI Agent Rediscover a Blaschke-Curve Invariant?
Artificial Intelligence
Summary
Finding hidden mathematical patterns is hard, but this study explores how an AI can help rediscover them without being told what to look for. The AI was given many examples of special curves and their related lines, then tried out ideas to explain what patterns might hold. It ended up fitting a very accurate mathematical formula predicting new lines almost perfectly. However, the AI’s result did not clearly beat a simpler traditional method, so the study focuses on how to carefully distinguish guessing from proof. The authors suggest their work as a method to organize how AI helps in math research, noting limits on evidence and reproducibility.
What this means in practice
- •For mathematical software developers: Improve tools that assist mathematicians by distinguishing between hypothesis and proof in numerical conjecture generation.
Tested on one dataset.
Authors
Yunus E. Zeytuncu
Abstract
We study generalized Blaschke curves as a controlled environment for AI-assisted mathematical rediscovery. For one fixed degree-four Blaschke product, an agent receives numerical coordinates of the six pair-lines determined by each of 80 boundary configurations. The target theorem is withheld from the task instructions. The saved research log reports rejected geometric hypotheses and a homogeneous cubic fitted to polygon sides. Its frozen coefficients predict 480 lines from 80 unseen parameter values, with a recorded RMS scale-free residual of $8.88\times10^{-17}$. Discovery-set diagonals provide an out-of-fit consistency check, not a fully held-out test. A separate one-configuration run reports insufficient evidence for invariance. A post-review deterministic degree-search baseline also recovers the cubic, so the experiment does not establish an advantage over polynomial fitting. We present this single-instance case study as a protocol for separating conjecture, numerical validation, and proof, with explicit limitations concerning agent metadata, prior knowledge, and reproducibility.