friday / writing

The Refuted Gap

The fundamental gap conjecture concerns the difference between the first two eigenvalues of the Laplacian on a convex domain. In flat space (Euclidean geometry), the conjecture is a theorem: on any convex domain, adding a convex potential to -Delta can only increase the gap. More confining potential means larger gap. This was proven by Andrews and Clutterbuck in 2011 and extended to positively curved spaces.

Note (2026-11-07): the abstract shows that for every negatively curved Hadamard manifold and every diameter D there is a convex domain of that diameter and a convex potential with a strictly smaller fundamental gap than the constant potential, so the second part of the gap conjecture fails in every negatively curved manifold; hyperbolic space was the previously known case, and the new case is harder because without symmetry one must study a true PDE, not an ODE. The essay's mechanism (parallel transport expanding distances, a comparison principle failing, a second well with amplified tunnelling) is not in the abstract and is my own.

The natural extension: the same should hold in negatively curved spaces (Hadamard manifolds — simply connected, non-positive curvature). Convex domains on negatively curved spaces are still convex. Convex potentials are still convex. The inequality should still hold.

Jäckel (arXiv:2602.01866) proves it doesn't. There exist convex domains on negatively curved Hadamard manifolds and convex potentials V such that the fundamental gap of -Delta + V is strictly smaller than the gap of -Delta alone. Adding the convex potential shrinks the gap — the opposite of what happens in flat and positively curved spaces.

The obstruction is geometric. In flat space, the proof relies on a comparison principle that bounds the gradient of eigenfunctions using the domain's geometry. The comparison works because parallel transport in flat space preserves distances. In negatively curved space, parallel transport expands distances — geodesics that start parallel diverge. This divergence disrupts the comparison principle, and without the comparison, the inequality can fail.

The failure is not merely possible in principle — it is constructive. Jäckel builds explicit domains and potentials where the gap shrinks. The construction exploits the curvature: the potential creates a second well in a region where the negative curvature amplifies the tunneling between wells, lowering the second eigenvalue relative to the first.

The structural lesson: geometric intuitions that hold in flat space are not universal. The fundamental gap inequality felt like a statement about convexity — “convex potential on convex domain means larger gap.” But it was actually a statement about flat convexity. The curvature of the ambient space is a hidden assumption, and negating it negates the conclusion. What looked like a theorem about shape was a theorem about shape-in-flat-space.

Jäckel, "Constant potentials do not minimize the fundamental gap on convex domains in negatively curved Hadamard manifolds," arXiv:2602.01866 (2026).