Tileable Surfaces: Geometry, Topology & Graphics

• Tileable surfaces are surfaces that admit edge-to-edge decompositions into congruent or polygonal patches, uniting classical tessellation, smooth R³ embeddings, and periodic texture mapping.
• They bridge diverse fields by linking combinatorial topology, differential geometry, and computer graphics to support consistent, seamless repetition in various applications.
• Recent advances include rigorous classifications via Euler characteristic and Delaney–Dress symbols, as well as neural implicit techniques for synthesizing seamless textures.

Tileable surfaces are studied in several distinct but related senses. In geometry and topology, the term refers to surfaces that admit edge-to-edge decompositions into polygons or congruent patches; in a recent differential-geometric formulation, it denotes $C^k$-embedded surfaces in $\mathbb{R}^3$ that admit geometric tilings by finitely many congruence classes of tiles; and in computer graphics it denotes surfaces or material maps whose textures repeat periodically without visible seams (Brander et al., 15 Jul 2025). A surface tessellation, in the classical sense, is a subdivision of a geometric surface into non-overlapping polygons whose interiors are disjoint and whose union is the surface (Faraco, 2023). For tileable texture maps, periodicity is expressed by

$f(x+T_x,y)=f(x,y), \qquad f(x,y+T_y)=f(x,y),$

so that copies laid side by side exhibit no discontinuities (Zhou et al., 2022). A further specialized usage occurs in the theory of flippable tilings, where constant-curvature surfaces are tiled by black and white faces satisfying a local forward/backward edge rule that supports a global flip operation (Fillastre et al., 2010).

1. Terminology and foundational definitions

The broadest mathematical notion is a tessellation of a surface $S$: a subdivision into polygons meeting edge-to-edge, with disjoint interiors and union equal to $S$ (Faraco, 2023). Regular tilings are encoded by the Schläfli symbol $\{p,q\}$, meaning regular $p$-gons with $q$ meeting at each vertex; semi-regular tilings allow multiple regular prototiles but require the same vertex configuration everywhere (Faraco, 2023). In this classical setting, tileability is a property of the surface together with a constant-curvature geometry.

A more restrictive smooth notion was introduced for embedded surfaces in $\mathbb{R}^3$. There, a tileable surface is a closed, connected $C^k$-embedded surface, $\mathbb{R}^3$0, possibly with boundary, that can be decomposed into tiles congruent by ambient rigid motions in $\mathbb{R}^3$1 (Brander et al., 15 Jul 2025). The prototile is homeomorphic to a closed disk, distinct tiles intersect only in subsets of their boundaries, and each tile meets only finitely many others (Brander et al., 15 Jul 2025). This framework also distinguishes admissible tiles, rigid tiles, finite edge type, and oriented tilings.

In graphics, tileability is usually a periodicity constraint on a texture or spatially varying BRDF map rather than on the underlying surface topology. If a surface patch is parameterized by UV coordinates and the texture function is periodic in both coordinates, standard wrap modes produce seamless repetition (Paz et al., 2024). This meaning is operational rather than topological: the surface may be arbitrary, but the applied material must have boundary consistency under repetition (Zhou et al., 2022).

These usages are not interchangeable. A surface can be topologically triangulable without carrying a monohedral smooth tiling by congruent patches, and a UV-mapped surface can support a tileable texture even when its intrinsic geometry has no regular tessellation. The literature therefore treats “tileable surfaces” as a family of related concepts rather than a single invariant notion (Brander et al., 15 Jul 2025).

2. Constant-curvature tessellations and geometric classification

For orientable surfaces, the classical starting point is Euler characteristic. A closed orientable surface of genus $\mathbb{R}^3$2 has

$\mathbb{R}^3$3

and a cell decomposition with vertices $\mathbb{R}^3$4, edges $\mathbb{R}^3$5, and faces $\mathbb{R}^3$6 satisfies

$\mathbb{R}^3$7

For a regular tessellation $\mathbb{R}^3$8, the incidence relations $\mathbb{R}^3$9 and $f(x+T_x,y)=f(x,y), \qquad f(x,y+T_y)=f(x,y),$0 imply

$f(x+T_x,y)=f(x,y), \qquad f(x,y+T_y)=f(x,y),$1

which directly ties combinatorics to curvature sign (Faraco, 2023).

The familiar trichotomy is

$f(x+T_x,y)=f(x,y), \qquad f(x,y+T_y)=f(x,y),$2

with $f(x+T_x,y)=f(x,y), \qquad f(x,y+T_y)=f(x,y),$3 for spherical tilings, $f(x+T_x,y)=f(x,y), \qquad f(x,y+T_y)=f(x,y),$4 for Euclidean tilings, and $f(x+T_x,y)=f(x,y), \qquad f(x,y+T_y)=f(x,y),$5 for hyperbolic tilings (Faraco, 2023). Thus the sphere supports only finitely many regular tilings with $f(x+T_x,y)=f(x,y), \qquad f(x,y+T_y)=f(x,y),$6, namely the Platonic families, while the torus inherits Euclidean periodic tilings such as $f(x+T_x,y)=f(x,y), \qquad f(x,y+T_y)=f(x,y),$7, $f(x+T_x,y)=f(x,y), \qquad f(x,y+T_y)=f(x,y),$8, and $f(x+T_x,y)=f(x,y), \qquad f(x,y+T_y)=f(x,y),$9, and higher-genus surfaces admit infinitely many hyperbolic regular tilings (Faraco, 2023).

Poincaré’s polygon theorem supplies a constructive criterion: a compact polygon $S$0 is a fundamental domain if its side pairings are orientation-preserving isometries and the vertex cycles have angle sums $S$1 for integers $S$2 (Faraco, 2023). In the hyperbolic case this produces closed surfaces by quotienting $S$3 by suitable Fuchsian groups. Conformally correct tilings refine this by preserving angles exactly on a compact embedded surface while allowing length distortion, which is necessary for hyperbolic patterns in $S$4 by Hilbert’s theorem (Schleimer et al., 2016).

Periodic tilings across all three geometries can be encoded combinatorially by Delaney–Dress symbols. In that language, Tegula enumerates $S$5 geometry-minimal periodic tilings of Dress complexity up to $S$6, with $S$7 spherical, $S$8 Euclidean, and $S$9 hyperbolic types (Zeller et al., 2020). This classification emphasizes that constant-curvature tileability is both a local angle problem and a global symmetry problem.

3. Flippable tilings, cone metrics, and polyhedral duality

Flippable tilings form a specialized theory on constant-curvature surfaces. A right flippable tiling $S$0 consists of black and white convex polygonal faces whose interiors are pairwise disjoint and cover the surface, together with geodesic edges satisfying a local orientation rule: along each oriented edge, the black face is forward on the right and backward on the left (Fillastre et al., 2010). The edge condition also requires equal lengths of the black intersections and, separately, of the white intersections along that edge (Fillastre et al., 2010).

The flip operation reverses the forward/backward status of black faces while preserving face shapes and combinatorics. On the round sphere, every right flippable tiling admits a unique left flippable tiling obtained by flipping, and flipping twice returns the original tiling (Fillastre et al., 2010). On a closed hyperbolic surface, a right flippable tiling likewise determines a unique hyperbolic metric and a left flippable tiling on that metric (Fillastre et al., 2010). Symmetric flippable tilings are those for which the flip preserves the constant-curvature structure.

A central tool is the passage from a tiling to black and white cone metrics obtained by gluing only black faces or only white faces. These satisfy the cone-metric Gauss–Bonnet relation

$S$1

where $S$2 are cone angles (Fillastre et al., 2010). In the hyperbolic case the black metric has cone angles $S$3, and the angle excess $S$4 equals the area of the corresponding white face; in the spherical case the cone angles satisfy $S$5, and $S$6 equals the area of the corresponding white face (Fillastre et al., 2010).

The theory is governed by three-dimensional duality. Spherical flippable tilings correspond to convex polyhedra in $S$7 and their polar duals, while hyperbolic symmetric flippable tilings correspond to Fuchsian equivariant convex polyhedral surfaces in $S$8 (Fillastre et al., 2010). Moduli spaces admit explicit dimension counts. For example, if $S$9 is a 3-connected incidence graph on $\{p,q\}$0 with $\{p,q\}$1 edges and $\{p,q\}$2 black faces, then $\{p,q\}$3 is non-empty and homeomorphic to an open ball of dimension $\{p,q\}$4, and globally $\{p,q\}$5 has dimension $\{p,q\}$6 (Fillastre et al., 2010). This places flippable tilings at the intersection of surface geometry, convexity, and moduli theory.

4. Smooth embedded tileable surfaces in $\{p,q\}$7

The recent smooth theory shifts attention from tilings on abstract constant-curvature surfaces to tilings by congruent patches on embedded surfaces in $\{p,q\}$8 (Brander et al., 15 Jul 2025). In this setting, tiles are closed sets with piecewise smooth boundaries, their interiors are homeomorphic to disks, and tangent continuity across shared edges is required to maintain $\{p,q\}$9 regularity of the ambient surface (Brander et al., 15 Jul 2025).

Several structural results are especially restrictive. If $p$0 is a complete $p$1-embedded $p$2-tileable surface with non-vanishing curvature, then $p$3 is a topological sphere (Brander et al., 15 Jul 2025). For compact $p$4 monotiled surfaces with Euler characteristic $p$5 and $p$6 tiles, the curvature contributed by the prototile satisfies

$p$7

while globally

$p$8

by Gauss–Bonnet (Brander et al., 15 Jul 2025). These formulas show that congruent tilings impose strong curvature budgets on the tile.

Rigidity sharpens the picture. A rigid prototile has a unique finite list of neighbors in any corona and unique rigid motions realizing the adjacencies; in that case every tile is rigid, each tile-to-tile isometry extends to an ambient rigid motion of the entire surface, and the resulting subgroup of $p$9 acts transitively on tiles (Brander et al., 15 Jul 2025). Finite edge type likewise imposes discrete angle-sum constraints at vertices: the sum of boundary interior angles contributed by all incident tiles must equal $q$0 (Brander et al., 15 Jul 2025).

The theory also exposes limitations. Not every monohedral polyhedron is smoothable into a finite edge type monotiled surface. The triaugmented triangular prism, with $q$1 equilateral triangular faces and vertex configuration $q$2, is proved not to be smoothable into a compact finite edge oriented monotiled surface with the same graph (Brander et al., 15 Jul 2025). Positive constructions exist, however, through deformations of planar and spherical tilings, lattice-periodic assemblies, and “pyramidal lifting” of regular tilings. This suggests that smooth tileable surfaces occupy a narrow but nontrivial region between polyhedral combinatorics and differential geometry.

5. Simplicial, Morse-theoretic, and shellable tileability

A different branch of the subject treats tileability for simplicial complexes. Here the basic $q$3-dimensional tiles are

$q$4

the complements of $q$5 facets in the standard simplex (Salepci et al., 2018). An $q$6-dimensional simplicial complex is tileable if its underlying space can be covered by pairwise disjoint $q$7-dimensional tiles $q$8, and a tiling has an $q$9-vector recording how many times each tile type occurs (Salepci et al., 2018). Skeletons and barycentric subdivisions preserve this kind of tileability (Salepci et al., 2018).

Morse tileability and Morse shellability broaden the framework by allowing Morse faces to be removed from simplex interiors. The central existence theorem for surfaces is that every triangulated closed surface is Morse shellable (Salepci et al., 2019). This does not imply that every triangulated closed surface is Morse tileable in the stricter sense; the paper explicitly notes that universal Morse tileability is not established (Salepci et al., 2019). Morse shellings encode compatible discrete Morse functions whose critical points correspond bijectively, with matching index, to the critical tiles (Salepci et al., 2019).

For orientable closed surfaces $\mathbb{R}^3$0, choosing a Morse shelling with one minimum and one maximum gives

$\mathbb{R}^3$1

since

$\mathbb{R}^3$2

forces $\mathbb{R}^3$3 (Welschinger, 2020). In dimensions less than four, products of closed manifolds admit triangulations with tame Morse shellings whose critical and $\mathbb{R}^3$4-vectors are palindromic (Welschinger, 2020). For products of orientable surfaces $\mathbb{R}^3$5, the critical vector is

$\mathbb{R}^3$6

obtained by convolution (Welschinger, 2020).

This simplicial viewpoint replaces geometric congruence by combinatorial decomposition and places tileability inside discrete Morse theory, shellability, and asymptotic subdivision theory. It also clarifies a common misconception: topological triangulability is universal for compact orientable surfaces, but strong monohedral or finite-edge-type tileability is not (Faraco, 2023).

6. Periodic materials, seamless textures, and tileability in graphics

In graphics, the most operational notion of a tileable surface is a surface carrying a material or texture that repeats seamlessly in UV space. TileGen addresses this for SVBRDFs by modifying StyleGAN2 so that all convolution, upsampling, and downsampling operations are wrap-around variants; indices are taken modulo spatial resolution, intermediate feature maps remain periodic at every level, and final outputs are smoothly tileable even when training data are not tileable (Zhou et al., 2022). The generator outputs diffuse albedo $\mathbb{R}^3$7, height $\mathbb{R}^3$8, roughness $\mathbb{R}^3$9, and, for metals, metallic $C^k$0, with shift consistency enforced by

$C^k$1

and inverse rendering from a single flash image formulated as

$C^k$2

in $C^k$3 space (Zhou et al., 2022).

Neural implicit representations provide a complementary approach. A sinusoidal INR with first-layer frequencies fixed to integer multiples of $C^k$4 produces only integer-frequency harmonics with period $C^k$5, so the learned texture is periodic by construction (Paz et al., 2024). The representation is continuous, directly evaluable at arbitrary coordinates, and can be regularized by a Poisson-inspired boundary term to improve seamlessness (Paz et al., 2024).

Evaluation has also become explicit. TexTile defines a differentiable, no-reference tileability score

$C^k$6

with $C^k$7 and $C^k$8, so the network directly “sees” concatenation artifacts (Rodriguez-Pardo et al., 2024). On a balanced test set, the final model achieves Error $C^k$9, Accuracy $\mathbb{R}^3$00, F1 $\mathbb{R}^3$01, and AUC $\mathbb{R}^3$02 (Rodriguez-Pardo et al., 2024). The score can guide optimization-based or diffusion-based synthesis toward more seamless textures (Rodriguez-Pardo et al., 2024).

Application pipelines use these ideas at scale. Plan2Scene synthesizes tileable textures for floors, walls, and ceilings from sparse, unaligned indoor photographs, then seam-corrects them before tiling across planar UV maps; on observed surfaces its Synth method reports Tile $\mathbb{R}^3$03 versus Crop’s Tile $\mathbb{R}^3$04 (Vidanapathirana et al., 2021). Content-aware tile generation via exterior boundary inpainting generalizes single self-tiling to Wang tiles and Dual Wang tiles using Stable-Diffusion-2-Inpainting, Euler sampling with $\mathbb{R}^3$05 inference steps, CFG scale $\mathbb{R}^3$06, and $\mathbb{R}^3$07 tiles (Sartor et al., 2024). Across these systems, seamless tileability is treated as periodic boundary consistency, but the practical target remains the same: surfaces that can be covered by repeated material maps without visible seams.

7. Specialized constructions, physical models, and current directions

Several specialized directions broaden the subject beyond standard tessellations. Conformally correct tilings preserve angles exactly on compact surfaces while allowing length distortion; a notable example is the Chmutov surface, tiled by hyperbolic $\mathbb{R}^3$08 triangles through conformal flattening and a binary search for the parameter $\mathbb{R}^3$09 (Schleimer et al., 2016). For hyperbolic surfaces, single-tile tilings form a finite problem when $\mathbb{R}^3$10: for an orientable surface of genus $\mathbb{R}^3$11, single-tile tilings require even $\mathbb{R}^3$12 in the range $\mathbb{R}^3$13, and the paper enumerates complete counts for small genera, including $\mathbb{R}^3$14 combinatorial tilings for genus $\mathbb{R}^3$15 at $\mathbb{R}^3$16 (Li et al., 27 Jan 2026).

Physical realizations add another layer. Curvagons are flexible regular polygon building blocks whose faces remain planar while curvature is concentrated at vertices; they realize Euclidean, spherical, hyperbolic, and mixed-curvature assemblies, with discrete Gauss–Bonnet expressed as

$\mathbb{R}^3$17

for the angle deficits $\mathbb{R}^3$18 (Kekkonen, 2022). At a very different dimensional scale, the unit $\mathbb{R}^3$19-ball can be tiled by $\mathbb{R}^3$20 congruent tiles, each congruent to a regular neighborhood of any chosen closed orientable surface smoothly embedded in $\mathbb{R}^3$21 (Ross et al., 13 May 2025).

Algorithmic and combinatorial variants persist. The hypercube is shown to have a face-unfolding that tiles $\mathbb{R}^3$22 and an edge-unfolding that tiles $\mathbb{R}^3$23, making it a dimension-descending tiler (Liang et al., 2015). Fault-free domino tileability on cylinders, tori, and Möbius strips admits complete classifications by parity and required-crosser arguments (Montelius, 2019). Tegula makes periodic tilings on $\mathbb{R}^3$24, $\mathbb{R}^3$25, and $\mathbb{R}^3$26 searchable and visualizable through Delaney–Dress symbols (Zeller et al., 2020).

Taken together, these directions show that tileable surfaces are not a single theorem but a landscape. At one end lie periodic materials and UV-space repetition; at another lie classical polygonal tessellations governed by curvature, Euler characteristic, and symmetry; at a third lie smooth embedded surfaces in $\mathbb{R}^3$27 constrained by rigidity, finite edge type, and Gauss–Bonnet. The unifying theme is local compatibility under repetition, but the precise meaning of “tileable” depends on whether the repeated object is a polygon, a smooth patch, a simplex, a BRDF map, or a regular neighborhood of a knotted surface.