Metric Triangulations in Geometry and Algorithms

• Metric triangulations are defined by intrinsic metric structures where distances, curvature, and scaling limits take precedence over combinatorial decompositions.
• They enable precise surface approximations and polyhedral metric constructions, including non-Euclidean Delaunay-type triangulations and intrinsic curvature computation.
• They extend to random triangulations and spaces of triangulations using bistellar moves and optimization techniques, linking discrete geometry with continuum limits.

Metric triangulations are triangulations studied through an explicitly metric structure rather than as purely combinatorial decompositions. In the literature represented here, this includes at least five closely related settings: approximating smooth surfaces by viewing triangulations as finite metric spaces with a Haussdorff-Gromov limit; constructing intrinsic polyhedral metrics by gluing Euclidean triangles edge-to-edge; treating random triangulations as metric-measure-curve objects with graph distance and scaling limits; defining Delaunay-type triangulations relative to non-Euclidean metrics such as Hilbert or anisotropic metrics; and endowing spaces of triangulations themselves with metrics such as bistellar-move distance or asymmetric Lipschitz-type distances on fixed triangulated Euclidean surfaces [0401023].

1. Principal meanings of the term

In the research record, “metric triangulations” does not designate a single formalism. It names a family of viewpoints in which metric data are primary and combinatorics are subordinate to intrinsic distance, curve length, graph distance, or move complexity.

Setting	Metric structure
Surface approximation	finite metric spaces with Haussdorff-Gromov limit
Metric surfaces	non-overlapping convex triangles of arbitrarily small diameter
Polyhedral surfaces	glued Euclidean triangles yielding an intrinsic metric
Random planar maps	graph distance, counting measure, and curve decoration
Spaces of triangulations	minimal number of bistellar transformations
Flip-graph optimization	Min Simplices, Min Diameter, Min Weight

This plurality matters because several common assumptions fail outside the Euclidean simplicial setting. A decomposition of a metric surface into convex triangles need not be a classical simplicial triangulation; a Delaunay triangulation in Hilbert geometry need not cover the Euclidean convex hull; and a “metric on triangulations” may refer either to an intrinsic metric carried by a single triangulated surface or to a metric on the configuration space of all triangulations (Creutz et al., 2021, Wu, 2023, Gezalyan et al., 2023, Lishak et al., 2016, Wang et al., 28 May 2026).

2. Intrinsic metric geometry on triangulated surfaces

A foundational metric program was stated in “Surface Triangulation -- The Metric Approach,” where approximating triangulations are treated as finite metric spaces and the target smooth surface as their Haussdorff-Gromov limit. In that formulation, principal directions, principal values, Gaussian curvature, and Mean curvature are to be defined in an intrinsic, discrete, metric way rather than by “approximating or paraphrasing the differentiable notions.” The first study concentrates on Gaussian curvature of a polyhedral surface via embedding curvature in the sense of Wald and Menger, and presents two modalities for employing these definitions in curvature computation [0401023].

A more general existence theorem for metric surfaces is given by “Triangulating metric surfaces,” which proves that any geodesic metric space homeomorphic to a closed surface may be decomposed into finitely many non-overlapping convex triangles, each of diameter at most an arbitrary prescribed $\varepsilon$. Its broader formulation covers length metric spaces homeomorphic to surfaces whose boundary components are piecewise geodesic, and produces a locally finite collection of non-overlapping triangles satisfying convexity relative to the boundary, diameter control, non-degeneracy, and a transit-point condition (Creutz et al., 2021). A central caveat is explicit: this is not a classical simplicial triangulation, because adjacent triangles are not required to meet edge-to-edge in the usual combinatorial sense. That caveat corrects a frequent misconception that every metric triangulation theorem automatically yields a simplicial complex structure.

The intrinsic polyhedral viewpoint is sharpened in “A characterization of the polyhedral metrics on triangulated surfaces.” Starting from a triangulated closed surface and a positive edge-length assignment satisfying triangle inequalities on each face, the paper constructs a polyhedral metric by gluing Euclidean triangles edge-to-edge and proves that the glued metric is the unique intrinsic metric on the surface that preserves the lengths of curves on the Euclidean triangles. It also develops edge-length coordinates on the Teichmüller space of polyhedral metrics on a marked surface, with the combinatorial identity $|E(T)| = 3|V| - 3\chi(S)$ for a triangulation with vertex set $V$ (Wu, 2023). Taken together, these works place metric triangulations within intrinsic geometry: the metric is not an auxiliary decoration but the defining object reconstructed from or decomposed by triangles.

3. Constant-curvature realizations and geometric triangulations

A major branch of the subject concerns triangulations compatible with ambient constant-curvature geometry. “Geometric moves relate geometric triangulations” defines a geometric triangulation of a Riemannian manifold as a finite triangulation in which the interior of each simplex is totally geodesic. For compact hyperbolic, spherical, or Euclidean manifolds, the paper proves that sufficiently many derived subdivisions of any two geometric triangulations are related by geometric bistellar moves; in dimensions $2$ and $3$, the triangulations are directly related by geometric bistellar moves without derived subdivision. In the spherical case, the diameter of the star of each simplex must be less than $\pi$ in the relevant statements (Kalelkar et al., 2019). The paper also emphasizes that not every combinatorial bistellar move is geometric, so metric compatibility constrains even local flip operations.

A related rigidity theorem appears in “Degree-regular triangulations of surfaces.” There, any degree-regular triangulation of a surface is shown to be geometric, meaning combinatorially equivalent to a geodesic triangulation with respect to a constant curvature metric on the surface. The classification separates the cases $d<6$, $d=6$, and $d>6$ into spherical, Euclidean, and hyperbolic behavior respectively, and the proof relies on the uniqueness of $d$-regular triangulations of the plane for $|E(T)| = 3|V| - 3\chi(S)$0 (Datta et al., 2017). This provides a purely combinatorial criterion forcing a constant-curvature metric realization.

The spherical case receives an arithmetic and moduli-theoretic treatment in “Triangulations of the Sphere.” Thurston’s construction starts from the Eisenstein lattice and produces triangulations of $|E(T)| = 3|V| - 3\chi(S)$1 in which every triangle is equilateral and every vertex is incident to $|E(T)| = 3|V| - 3\chi(S)$2 or $|E(T)| = 3|V| - 3\chi(S)$3 triangles. Realizing each $|E(T)| = 3|V| - 3\chi(S)$4-simplex as a flat equilateral triangle yields a piecewise Euclidean metric on the sphere that is flat except at cone points; when every singularity comes from a $|E(T)| = 3|V| - 3\chi(S)$5-valent vertex, there are exactly $|E(T)| = 3|V| - 3\chi(S)$6 cone points, each with angle deficit $|E(T)| = 3|V| - 3\chi(S)$7, by the discrete Gauss–Bonnet theorem. The paper describes the moduli space $|E(T)| = 3|V| - 3\chi(S)$8 of such metrics modulo rescaling as open and dense in an orbifold $|E(T)| = 3|V| - 3\chi(S)$9, and interprets the compactification as allowing collisions of cone points (Baez, 8 Jun 2026). In this sense, metric triangulations become coordinates on a moduli problem for flat cone metrics on the sphere.

4. Random triangulations as metric-measure surfaces

In random geometry, a triangulation is frequently treated as an intrinsic metric-measure-curve object. “Joint scaling limit of site percolation on random triangulations in the metric and peanosphere sense” studies Boltzmann triangulations with Bernoulli-$V$0 site percolation, viewed through the graph distance, the counting measure on vertices, and percolation interfaces. The paper proves joint convergence of the space-filling exploration and its walk encoding: after scaling, the decorated triangulation converges in Gromov–Hausdorff–Prokhorov–uniform topology to the $V$1-Liouville quantum gravity disk, equivalently the Brownian disk, decorated by space-filling SLE$V$2, while the discrete two-dimensional boundary-length walk converges to the correlated Brownian motion of the mating-of-trees theorem. It also upgrades metric convergence from a single interface to the full loop ensemble, obtaining CLE$V$3 on the Brownian disk (Gwynne et al., 2019). The metric scaling relations are part of the statement: distances scale like $V$4, measures like $V$5, and boundary lengths like $V$6.

“Convergence of uniform triangulations under the Cardy embedding” strengthens the metric viewpoint by embedding a random triangulation into an equilateral triangle $V$7 via percolation crossing probabilities. The induced metric on $V$8 is defined by pulling back the rescaled graph distance through the nearest-vertex map, while the area and boundary measures are pushforwards of vertex-counting measures. The main theorem states that, for uniformly sampled triangulations with boundary length of order $V$9, the metric and measures converge jointly to those of the $2$0-LQG disk, equivalently the Brownian disk, conformally embedded into $2$1 (Holden et al., 2019). These results show that a triangulation can carry enough intrinsic metric information to converge not merely as a graph or map, but as a continuum random surface.

5. Metric-dependent Delaunay structures beyond Euclidean geometry

Another important meaning of metric triangulation arises when the notion of “Delaunay” is defined relative to a non-Euclidean metric. “Delaunay Triangulations in the Hilbert Metric” adapts Euclidean Delaunay theory to the Hilbert geometry defined by a convex polygon in the plane. The Hilbert Delaunay triangulation is defined as the dual of the Hilbert Voronoi diagram, and although it remains a planar straight-line graph that spans the site set and contains the Hilbert MST and Hilbert RNG, it has a decisive non-Euclidean feature: it does not necessarily cover the convex hull of the point set. The paper therefore introduces the Hilbert hull, defined as the region covered by the Hilbert Delaunay triangulation, as well as Hilbert balls at infinity to capture boundary effects. It presents a randomized incremental algorithm with expected running time $2$2 for $2$3 sites in the Hilbert geometry of a convex $2$4-gon, and a Hilbert hull algorithm running in $2$5, where $2$6 is the number of points on the hull boundary (Gezalyan et al., 2023). A standard Euclidean intuition is therefore false in this setting: a Delaunay triangulation need not triangulate the Euclidean convex hull.

A different metric deformation of Delaunay theory appears in “Duals of Orphan-Free Anisotropic Voronoi Diagrams are Triangulations.” Working in $2$7 with a continuous Riemannian metric field $2$8, the paper uses the anisotropic distance

$2$9

to define anisotropic Voronoi regions. Under the orphan-free condition, meaning each Voronoi region is connected, the dual is an embedded polygonal mesh with convex faces; under bounded anisotropy, it triangulates the convex hull of the sites and has properties parallel to ordinary Delaunay triangulations. A central geometric substitute for the empty circumcircle property is the empty circum-ellipse property (Canas et al., 2011). These results place metric triangulations within mesh generation and computational geometry, where the ambient metric controls both admissible dual structure and global embedding behavior.

6. Metrics on spaces of triangulations and on fixed triangulated structures

The metric viewpoint also turns from individual triangulations to spaces of triangulations. “Sizes of spaces of triangulations of 4-manifolds and balanced presentations of the trivial group” defines $3$0 for a compact $3$1-dimensional PL-manifold $3$2 as the set of simplicial isomorphism classes of triangulations of $3$3, endowed with the metric given by the minimal number of bistellar transformations needed to pass from one triangulation to another. The paper proves that there exists $3$4 such that, for each positive integer $3$5 and all sufficiently large $3$6, there are more than $3$7 triangulations of $3$8 with fewer than $3$9 simplices whose pairwise distances are at least $\pi$0, where $\pi$1 (Lishak et al., 2016). Here the metric measures transition complexity in a discrete configuration space rather than intrinsic surface geometry.

“TriSearch: Learning to Optimize Triangulations via Bistellar Flips” uses essentially the same flip-graph viewpoint algorithmically. A triangulation of a polytope is a node of the flip graph, feasible actions are flippable circuits, and the objective is to optimize a function $\pi$2 over triangulations. The metric objectives in the experiments are Min Simplices, Min Diameter, and Min Weight. The framework uses a circuit-supported subtriangulation action representation, casts search as an MDP, and reports zero-shot generalization from small training instances to larger polytopes with exponentially larger search spaces. In $\pi$3D it achieves an average relative gap of $\pi$4 compared with $\pi$5 for the next-best baseline, and in $\pi$6D an average relative gap of $\pi$7 versus $\pi$8 for the next-best baseline (Wang et al., 28 May 2026). In this usage, “metric triangulations” refers to optimization of quantitative geometric and combinatorial objectives over triangulation space.

A more continuous metric geometry of triangulated structures is developed in “On spaces of Euclidean triangles and triangulated Euclidean surfaces.” The paper introduces an asymmetric metric on the space of marked Euclidean triangles, proves equivalent descriptions via coordinate ratios and best Lipschitz maps, describes geodesics, shows the metric is Finsler, and extends the theory to convex Euclidean polygons and to surfaces equipped with singular Euclidean structures with a fixed triangulation. For a triangulated Euclidean surface with face index set $\pi$9, the metric is

$d<6$0

The resulting spaces admit a completeness theory adapted to asymmetric metrics, and in the fixed-triangulation surface setting the relevant metric spaces are complete (Saglam et al., 24 Apr 2025). This suggests a final, structurally important distinction: a metric triangulation may be either a triangulation endowed with an intrinsic metric or a point in a metric space of triangulated geometric structures.

Across these lines of work, metric triangulations form a unifying theme rather than a single definition. The unifying principle is that triangles are not merely combinatorial cells: they are carriers of intrinsic distance, curvature, conformal data, non-Euclidean proximity, or move complexity. That principle supports problems ranging from curvature computation on polyhedral surfaces to Brownian scaling limits, non-Euclidean Delaunay theory, arithmetic moduli of cone metrics, and algorithmic navigation of vast flip graphs.