Prince Rupert of the Rhine predicted in the 17th century that a cube has an interesting spatial property: you can cut a tunnel through a cube large enough to pass an identical cube through it. John Wallis proved him right. Later mathematicians showed that all regular polyhedra and many other convex shapes share this property — dubbed Rupert's property. A 2017 conjecture proposed that all convex 3D bodies are Rupert.
Jakob Steininger and Sergey Yurkevich disproved the conjecture by constructing the noperthedron — a convex polyhedron with 90 vertices, 240 edges, and 152 faces that cannot pass through a hole in itself. No matter how you orient the second copy, no straight tunnel through the first copy accommodates it.
The proof method: divide the parameter space of orientations into approximately 18 million tiny blocks. Check the center point of each block to determine whether that orientation produces a passage. None does. The parameter space is fully covered. The conjecture is false.
The structural observation: the proof establishes that the conjecture is false without explaining why. The 18 million checked orientations don't reveal which geometric feature of the noperthedron prevents passage. They demonstrate THAT every orientation fails, not what makes the shape resistant. The proof is a verification, not an explanation.
This is a recurring pattern in computational mathematics. The four-color theorem (1976) proved that four colors suffice for any planar map by checking 1,936 reducible configurations. The proof is correct. It offers no insight into why four colors work — no structural principle that a human could hold in their mind and use to understand coloring generally. Similarly, the noperthedron proof offers no principle for identifying non-Rupert shapes. If a simpler shape without Rupert's property exists, the exhaustive search won't find it — because the search was tailored to the noperthedron's specific geometry.
Computational proofs trade understanding for certainty. A traditional proof earns certainty through understanding: you see why the theorem holds, and the “why” is what convinces you. A computational proof earns certainty through coverage: you check every case, and the completeness is what convinces you. Both produce knowledge that the theorem is true. Only one produces knowledge of why.