The longest-edge bisection algorithm may produce degenerating tetrahedra

Technical Summary: "The longest-edge bisection algorithm may produce degenerating tetrahedra"

Problem Statement
The paper addresses a critical gap in the theory of tetrahedral mesh refinement. While the longest-edge bisection algorithm is a standard technique for constructing nested simplicial meshes, its behavior in three dimensions is not fully understood regarding nondegeneration. In two dimensions, the convergence and shape regularity of triangles generated by repeated longest-edge bisection are well-established. However, in three dimensions, the unrestricted geometric instruction—where one must bisect a currently longest edge but may select any edge if multiple edges are tied in length—lacks a guarantee of nondegeneration. Previous studies have focused on marked-edge schemes (which ensure finite similarity classes) or specific tetrahedral families, but the behavior of the general, unrestricted rule with arbitrary tie-breaking remains an open question.

Methodology
The author constructs an explicit, exact counterexample to demonstrate that the unrestricted rule can lead to degeneration. The methodology involves:

1. Defining a Parametric Family: A specific tetrahedron $E(a)$ is defined with vertices dependent on a parameter $0 < a \le 1$.
2. Two-Step Recurrence: The paper analyzes a two-step bisection process:
   • Step 1: Bisect the unique longest edge $AD$ of $E(a)$ and retain the child tetrahedron $O(a)$.
   • Step 2: In $O(a)$, the edges $BD$ and $DC$ are exactly tied for the longest length. The author selects $BD$ for bisection, retains the resulting child $F(a)$, and relabels its vertices.
3. Exact Congruence Proof: It is proven that the resulting tetrahedron $F(a)$ is congruent to the original family member $E(a/2)$. This establishes a recurrence relation where the shape parameter $a$ is halved at every two bisection steps.
4. Iterative Construction: Starting with $E(1)$, this process generates an infinite sequence of tetrahedra $E(1) \to O(1) \to E(1/2) \to O(1/2) \to \dots$.
5. Regularity Analysis: The author calculates the normalized volume ratio ($|T|/h_T^3$) and specific dihedral angles for the sequence as $a \to 0$ (where $a_k = 2^{-k}$) to test against standard regularity criteria.

Key Contributions and Results
The paper provides a rigorous proof that the unrestricted longest-edge bisection algorithm in 3D can produce a sequence of degenerating tetrahedra. The specific findings are:

• Violation of Shape Regularity: The normalized volume ratio $|E_k|/h_{E_k}^3$ tends to zero as $k \to \infty$. Specifically, the ratio decays asymptotically as $2^{-k/2}$, proving that no uniform positive lower bound exists for the family.
• Violation of Minimum-Angle Condition: The interior dihedral angle $\delta(a)$ at edge $CD$ in $E(a)$ tends to zero. The paper shows $\delta(a) \sim \sqrt{8a/7}$, meaning the sequence contains arbitrarily "flat" angles.
• Violation of Maximum-Angle Condition: The interior dihedral angle $\Theta(a)$ at edge $MC$ in $O(a)$ tends to $\pi$. The cosine of this angle approaches $-1$, indicating the tetrahedra become arbitrarily "sliver-like" or flat in a different configuration.
• Role of Tie-Breaking: The degeneration is driven by a recurrent tie in the longest edge lengths ($BD = DC$). The paper demonstrates that a simple deterministic tie-breaking rule (choosing the edge opposite the longer edge) is sufficient to select this "bad" branch.

Significance and Claims
The paper's primary claim is modest but definitive: it proves the existence of a degenerating admissible orbit under the unrestricted longest-edge rule. The title's use of "may" is emphasized as essential; the construction does not assert that every tie-breaking convention leads to degeneration, nor does it claim that all orbits are degenerate. Rather, it establishes that without explicit tie-resolution mechanisms included in the algorithm and analyzed as part of the regularity theorem, the algorithm is not guaranteed to produce nondegenerate meshes.

The work serves as a counterexample to the assumption that the geometric selection rule alone is sufficient for 3D regularity. It highlights that in three dimensions, the behavior of refinement depends critically on the marking and tie-breaking conventions, and that arbitrary choices among tied longest edges can lead to a loss of shape regularity, minimum-angle bounds, and maximum-angle bounds simultaneously.