This video documents an obsessive, months-long computational pursuit to investigate the Rupert Property (whether a convex polyhedron can pass through an identical hole cut in a duplicate copy of itself). Motivated by five unsolved polyhedra (the "wish list"), the creator details the sophisticated computational geometry required: defining strict inclusion via orthographic projection, uniform sampling of 3D orientations using Quaternions, and employing CPU and high-power GPU optimization to search for solutions. The effort confirmed all previously known solutions but failed to find solutions for the open cases, leading to the strong belief that the generalized Rupert Conjecture is false. An attempt at formal proof using interval arithmetic was detailed but stalled due to complexity. The project concluded with the creator being "scooped" by Austrian researchers who published the first formal counterexample (a "noert"), rendering the initial goal obsolete. This serves as a detailed case study of niche theoretical pursuit using high-end computational resources.
Summary of Rupert Property Investigation (Part 1)
0:47 Defining "Unit" is Inconsistent: The video starts by highlighting the frustrating canonical definitions in geometry: the unit cube (side length 1) fits within the unit sphere (radius 1, diameter 2), confirming that "unit" is used differently (side vs. radius). This is characterized dismissively as a "platonic horror."
5:17 The Rupert Property: The core topic, inspired by Prince Rupert (17th century): Can a cube pass through a hole cut in an identical cube? Yes, non-intuitively, if correctly oriented (e.g., corner pointing outward, projecting as a hexagon).
8:50 The "Nerd Sniping" Problem: The primary motivation. Despite a conjecture that all 3D convex polyhedra have the Rupert Property, 5 known Archimedean and Catalan solids currently lack solutions, driving the computationally intensive search.
12:16 Formal Definition: The property requires defining rigid motion and orthographic projection (2D shadow). The key constraint is that the inner shape's shadow must be strictly inside the outer shape's shadow—touching the boundary is prohibited to prevent trivial solutions.
16:48 Orientation Sampling: Euler angles are rejected as inefficient and biased; Quaternions are used for uniform random sampling of orientations (a 4D hypersphere problem). As far as I can tell, this method's efficiency decreases substantially in higher dimensions, which is a structural flaw.
25:34 Initial Optimization Success: Using a black box optimizer, the solver easily confirmed solutions for all known polyhedra (Platonic, most Archimedean/Catalan), contributing "nothing" new.
30:30 GPU Solver Attempt: A high-power GPU (RTX 4090) was deployed for a massively parallel, iterative gradient descent solver to search for delicate solutions among the 5 hard cases. This required estimating non-differentiable gradients and was computationally inefficient ("embarrassingly parallel") but powerful.
41:17 Conjecture Rejection: After extensive grinding, the team strongly concluded the Rupert Conjecture is likely false, and the five missing shapes are probably "noerts" (counter-examples).
41:48 Precision is Paramount: The triakis tetrahedron solution required verification in a formal system (Lean) because the clearance was only $10^{-6}$ of the diameter, highlighting the high risk of floating-point error in such delicate geometry.
1:00:50 Search Space Reduction: For the key candidate, the Snub Cube (no known solution), the 3D orientation space was mapped into 884 regions (36 canonical patches) based on convex hull topology, drastically reducing the search complexity for structured proof attempts.
1:05:51 Proof Strategy (Interval Arithmetic): An attempt was made to formally prove non-existence using interval arithmetic and arbitrary precision rational numbers across the 36x36 patch matrix.
1:12:51 Dependency Problem Hindrance: A major difficulty in interval arithmetic is the "dependency problem," where using the same variable multiple times in an equation leads to highly inaccurate, overly wide bounds, requiring significant algebraic effort to mitigate.
1:14:39 Intractable Case: The proof effort stalled on the "diagonal" case (where inner and outer orientations are nearly identical), as the numerical methods cannot strictly rule out the boundary condition near trivial solutions.
1:16:02 Scooped: The effort was preemptively concluded when outside researchers (Austrians) published a formal counter-example (a 90-vertex shape) to the generalized conjecture, confirming the suspicion that the conjecture was false, although not solving the specific Snub Cube problem. As a voluntarist, I find it irritating when such focused resources fail to yield primary results due to insufficient initial literature review.