← All papers
First page of Balancing non-rectangular tables

Balancing non-rectangular tables

Antoine Gournay

math.HO Apr 2, 2020 · v3
Nearly all results on stabilising wobbly cyclic-quadrilateral tables, including computer-assisted theorems, are formalised in Lean, with code in a public repository.
Balancing square and rectangular tables by rotation has been an interesting way to illustrate the intermediate value theorem. This article studies the same question for tables whose feet form a cyclic quadrilateral, following a conjecture of A. Martin. In the idealised model behind this conjecture, the balancing act works for every such table and is a nice application of the ergodic theorem. For the actual rigid table on a Lipschitz terrain, the situation is more subtle. If the terrain is flattened enough, the table can be stabilised unless an idealised wobbling function vanishes identically, \ie unless the terrain is degenerate for the table. Even if this should happen, the tables whose feet are vertices of a regular polygon and which satisfy a non-degeneracy condition of their own can still be stabilised on every terrain with a small enough Lipschitz constant (\eg a halved regular hexagon as long as the slope is less than $3.7$\textdegree{}). On the other hand, there is an isosceles trapezoid and another table, which are shown to wobble at every angle on suitable terrains however small the Lipschitz constant is. These counterexamples are fragile: for every table, if the idealised wobbling function vanishes identically for every position of the rotation axis in a small disc, then, near the feet, the terrain is a plane or a paraboloid of revolution with a vertical axis; so in practice a slight move of the axis makes the terrain non-degenerate. The results are, with minor exceptions, formalised in Lean.

Square and rectangular tables on uneven ground can be stabilised by rotation. The question is whether this also works for tables whose feet form a cyclic quadrilateral, as conjectured by A. Martin, when the table is rigid and the terrain is Lipschitz.

In an idealised model where the feet are put down vertically, the ergodic theorem shows balancing is always possible. For the rigid table, an 'equal hovering position' is shown to be well-defined and continuous for small Lipschitz constants, and its wobbling function is compared with the idealised one up to a third-order error. Degenerate terrains are then analysed separately for regular-polygon tables and for counterexamples. The results, including computer-assisted theorems, are formalised in Lean, with AI assistance from Claude and Aristotle.

If the terrain is flattened enough, the table can be stabilised unless the idealised wobbling function vanishes identically. Tables cut from regular polygons that satisfy a non-degeneracy condition stabilise on every terrain with a small enough Lipschitz constant. An isosceles trapezoid and another table can wobble at every angle on suitable terrains, but these counterexamples are fragile under small moves of the rotation axis. All statements of §4–5 are formalised in Lean, except numerical observations.