This video chronicles the speaker's multi-month "math hole" delving into the Rupert Property, which investigates whether a convex polyhedron can pass through another identical copy of itself. The speaker introduces the historical problem (e.g., Prince Rupert's cube) and highlights a persistent conjecture that all convex polyhedra possess this property, despite a few well-known Archimedean and Catalan solids lacking a confirmed solution. Motivated by this "nerd sniping," the speaker details his computational efforts, employing GPU-accelerated solvers, blackbox optimizers, and a meticulous exploration of the solution space using "patches" of hull topologies. He also explores various methods for generating "noperts"—polyhedra believed to be counter-examples—and statistical analysis of prisms. The video culminates with the speaker's development of rigorous interval arithmetic techniques to prove the absence of solutions for certain cases, only to be "scooped" by other researchers who formally demonstrated the existence of a nopert. Despite this, the video maintains a focus on the snub cube as a compelling, yet unproven, counter-example, underscoring the ongoing challenges in computational geometry.
Exploring the Rupert Property: A Dive into Polyhedra and Computational Geometry
0:03 The "Math Hole": The speaker recounts getting caught in a multi-month "math hole" trying to solve an unsolved problem found online.
0:43 The Rupert Property Introduced: The core problem is presented: can a unit cube (or any identical polyhedron) pass through another unit cube? It can, which is surprising. This is characterized as a "Platonic Horror"—a beautiful problem with an upsetting solution.
3:12 Prince Rupert's Question: The origin of the problem is attributed to Prince Rupert of the Rhine, who posed whether a cube could pass through another cube of the same size.
5:37 Platonic Solids and Duality: The video discusses the five Platonic Solids (tetrahedron, cube, octahedron, dodecahedron, icosahedron), their dual relationships, and why only these five exist.
8:49 The Unsolved Cases: It's noted that while the Rupert property is conjectured for all 3D convex polyhedra, 3 of the 13 Archimedean solids and 2 of the 13 Catalan solids lack known solutions, prompting the speaker's research.
11:36 Formal Definition: The Rupert Property is formally defined: two identical convex shapes must be orientable and projectable such that the inner shape's 2D shadow fits strictly inside the outer shape's shadow along a given axis.
16:22 Efficient Orientation Sampling: Quaternions are used over Euler angles for uniformly sampling 3D orientations due to Euler angles' biases and gimbal lock issues. Rejection sampling is employed to generate random unit quaternions.
22:21 Convex Hull & Blackbox Optimization: The 2D shadow of a polyhedron is computed using the convex hull algorithm. A blackbox optimizer with a custom loss function is then used to find configurations where the inner shadow fits inside the outer, demonstrating solutions for the dodecahedron and rhombicuboctahedron.
30:03 GPU-Accelerated Solver: A GPU-based solver is developed using gradient descent and "embarrassingly parallel" triangulation to speed up the search for solutions, albeit with some numerical approximations.
40:02 Believing the Conjecture is False & "Noperts": The speaker and his collaborators (David and Jason) come to believe the Rupert conjecture (all polyhedra have the property) is false. They coin the term "noperts" for polyhedra that do not possess the Rupert property.
41:40 Delicate Solutions: Some known solutions exist with extremely small clearances ("hanging by a thread"), which David formally verified using Lean to mitigate floating-point errors.
43:18 Searching for Noperts: Adversarial methods and deoptimization procedures are used to generate polyhedra that are difficult to solve, aiming to find actual noperts.
45:16 Upsetting Truths about Symmetry: The video explores 3D symmetries, revealing an "upsetting" mathematical truth that in 5D and higher dimensions, only two irreducible rotational symmetries exist (tetrahedral and octahedral), unlike the richer symmetries of 3D and 4D.
53:40 Candidate Noperts Identified: Two potential "noperts" are presented: Nopert #76 (20 vertices, 5-way cyclic symmetry, triangular faces) and Nopert #214 (20 vertices, 27 faces, with some quadrilateral faces). These remain unsolved by the author's extensive search, but lack formal proof.
55:41 Prisms Analysis: Statistical analysis of prisms shows they are likely all solvable, with distinct "manhole" (shallow) and "churro" (long) solution types, and complex chaotic solution spaces in between.
59:50 Focusing on the Snub Cube: The snub cube (an Archimedean solid with 24 vertices) becomes the main focus due to its canonical status and symmetries. Its view directions are categorized into 36 canonical "patches" for detailed analysis.
1:05:18 Interval Arithmetic for Proof: To formally prove a polyhedron is a nopert, the author employs interval arithmetic combined with a bisection method across the 7-dimensional parameter space to rigorously rule out all possible solution configurations. Challenges include the "dependency problem" and handling the "diagonal" where inner and outer shapes are almost identical.
1:15:43 Scooped!: The author reveals he was "scooped" by other researchers who published a paper proving the existence of a nopert—their "noered hedin" with 90 vertices. Their approach was similar, but their methods don't directly apply to proving the snub cube is a nopert, requiring a "new local theorem."
1:18:03 Continued Intrigue: Despite being scooped, the speaker maintains a humorous outlook, implying the mathematical "math hole" continues, even if the primary goal has been achieved by others.