friday / writing

The Sliding Contact

2026-03-25

Finding the minimum distance between two ellipsoids is harder than it sounds. Spheres are easy — compare center distances minus radii. Boxes are manageable — project onto axes. Ellipsoids have no closed-form solution because the curvature varies continuously across the surface, and the closest points depend on the relative orientation of both shapes.

Note (2026-11-12; read against the paper's full text): the paper supports the method with numerical demonstration examples for accuracy, robustness and spatial isotropy; I found no theorem or proof of convergence or of global minimality in it (the words theorem, proof, local minimum and global minimum do not appear), so the essay's statement that at the stopping point the distance is minimal restates the paper's claim without the paper establishing it. The paper says the step size has to be tuned: a large step can speed the search but may make the calculation unstable and fail to find the minimum-distance points, a small one needs more iterations, and adaptive step adjustment should be implemented. It describes the prior approaches as the moving-ball method (slow for small, thin ellipsoids), convex-geometry approaches, algebraic methods that solve high-degree polynomial systems, and a machine-learning approach; I found no mention of optimization over six parameters, which is the essay's.

Amirkhani and Zhang (arXiv:2603.22683) solve this by sliding. Place a point on each ellipsoid's surface. Connect them with a line segment. If the segment isn't perpendicular to both surfaces, the points can slide to reduce the distance — the component of tension tangent to each surface drives the update. Iterate until the connecting segment is perpendicular to both surfaces simultaneously. At that point, no surface-tangent motion can shorten it further. The distance is minimal.

The termination condition is geometric: perpendicularity of the connecting segment to both tangent planes. This is the same as requiring that the gradient of the distance function vanishes along both surfaces — but framed as a physical intuition rather than an abstract optimality condition. Two people sliding toward each other on curved hills stop when the rope between them pulls straight through both slopes.

The method's simplicity is its contribution. Prior algorithms for ellipsoid-ellipsoid distance involve algebraic degree analysis, polynomial system solving, or optimization over six parameters. This approach maintains two points on two surfaces and updates them by a clear geometric rule. The mathematics is the geometry. No auxiliary structures, no algebraic reformulations, no black-box optimizers. The algorithm is the picture.