What Happens When a Fields Medalist Teams Up with ChatGPT‑5? An AI‑Assisted Geometry Case Study

Problem Statement

Consider a smooth immersed sphere \(\Sigma\) in three‑dimensional Euclidean space \(\mathbb{R}^3\) whose two principal curvatures satisfy \(|k_1|,|k_2|\le 1\). The open question asks whether the volume \(V(\Sigma)\) enclosed by \(\Sigma\) is always at least the volume of the unit sphere, i.e. \(V(\Sigma)\ge \frac{4\pi}{3}\).

Analytical Decomposition

Perturbative regime: \(\Sigma\) is close to the standard unit sphere (small deviation in geometry).

Non‑perturbative regime: \(\Sigma\) may differ significantly from a round sphere.

The investigation focused first on the perturbative regime, where classical analytic tools are most effective, and later extended to the more general star‑shaped case.

AI‑Assisted Computations

Using a language model, the required geometric quantities for a star‑shaped surface were computed explicitly. The AI combined several integral identities:

Stokes’ theorem to convert surface integrals into volume integrals.

Willmore inequality and the Gauss–Bonnet theorem to control total curvature.

Minkowski’s first integral formula to relate curvature integrals to the support function of a star‑shaped body.

By inserting these relations, the AI produced a concise proof that the unit sphere minimizes volume under the curvature bound. The core argument can be summarized as follows:

∫_Σ (1 - H) dA = 0  ⇒  H ≡ 1  ⇒  Σ is a round sphere

where \(H\) is the mean curvature. Equality forces \(Σ\) to be the unit sphere, establishing the desired volume inequality.

In addition, the AI generated two independent proof variants:

Using the divergence theorem to relate the volume to an integral of the radial component of the position vector.

Employing a geometric flow (e.g., inverse mean curvature flow) that preserves the curvature bound while monotonically increasing volume, leading to the same extremal conclusion.

Limitations and Errors

During the perturbative analysis, the AI introduced a small algebraic slip in estimating a nonlinear term; the error is comparable to a typical mistake made by a nonlinear PDE specialist.

The AI did not flag an incorrect assumption made by the researcher that the immersed sphere’s inner radius is uniformly bounded, thus offering no corrective strategic guidance at the medium‑scale level.

Further Directions Suggested by AI

The model proposed treating the “almost round” situation as a perturbative elliptic PDE problem. By linearizing the curvature condition around the unit sphere, one obtains an elliptic operator \(L\) acting on the radial perturbation \(u\): L u = Δ_{S^2} u + 2u + O(u^2) Standard elliptic regularity and coercive estimates then yield control of \(u\) in Sobolev norms, suggesting a pathway to a rigorous proof for sufficiently small deviations. The AI also outlined a brute‑force numerical scheme that discretizes the surface and searches over admissible star‑shaped configurations, but warned that such an approach lacks theoretical insight.

Related Work

The two‑dimensional analogue of the problem is resolved by the Pestov–Ionin theorem. For the three‑dimensional star‑shaped case, relevant analyses appear in papers by Pankrashkin and by Qiu, which develop integral‑geometric techniques similar to those employed by the AI.

Reference: https://mathstodon.xyz/@tao