🔢 mathematics

Polyhedral design with blended $n$ -sided interpolants

This paper proposes a new parametric surface representation that interpolates vertices of arbitrary closed meshes by smoothly blending local, multi-sided quadratic interpolants using special rational curve parameterizations for non-four-sided cases.

Original authors: Péter Salvi

Published 2026-01-28

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Péter Salvi

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are an architect trying to build a smooth, seamless roof over a complex, irregularly shaped building. Your blueprint isn't a perfect grid; it's a mesh of polygons with different numbers of sides—some squares, some triangles, some pentagons. The challenge is to cover this entire structure with a single, smooth skin that touches every corner of your blueprint exactly, without any bumps, cracks, or awkward seams where the pieces meet.

This paper by P´eter Salvi proposes a new "digital skin" for such 3D models. Here is how it works, broken down into simple concepts:

The Core Idea: Blending Local Patches

Instead of trying to stretch one giant, perfect sheet over the whole shape (which is mathematically impossible for irregular shapes), the author suggests building the surface out of many small, overlapping patches.

Think of it like laying down a floor with tiles. Usually, you use square tiles. But if your room has a weird corner, you might need a triangular or pentagonal tile. The problem is, if you just glue square tiles next to a pentagonal one, you get jagged edges.

Salvi's method is like using smart, flexible tiles that can stretch and blend into their neighbors perfectly.

The Input: You start with a "cage" (a wireframe mesh) of any shape.
The Local View: Around every single corner (vertex) of this cage, the computer looks at the immediate neighbors to create a small, local "net" or control system.
The Blend: It creates four small surface pieces around that corner and blends them together like mixing paints, ensuring the transition is smooth.

Handling the "Odd Ones Out"

Most computer graphics methods work great on perfect squares (quadrilaterals). But real-world objects often have "irregular" spots—corners where 3, 5, or 6 lines meet instead of 4.

The Square Case: If the mesh is made of perfect squares, the math is straightforward. It uses standard quadratic curves (like the arc of a rainbow) to connect the dots.
The Irregular Case: When the mesh has a corner with 5 sides (a pentagon) or 3 sides (a triangle), standard math breaks down. The author introduces a special technique using rational curves.
- Analogy: Imagine trying to draw a smooth line from the center of a star to its points. If you use a straight ruler, it looks stiff. If you use a flexible rubber band that is weighted differently at different points, you can pull it into a perfect curve that fits the star's shape. The paper uses a similar "weighted" mathematical trick to map the square computer grid onto these weird, multi-sided shapes without tearing the surface.

The "Magic" of Interpolation

A key feature of this method is interpolation.

Approximation (The old way): Many 3D tools create a surface that looks like the shape but floats slightly above or below the actual points, smoothing them out like a heavy blanket.
Interpolation (This paper's way): This method forces the surface to touch every single point of the original wireframe exactly. It's like a tight-fitting glove that hugs every finger perfectly, rather than a loose sock. The author argues that because the surface is forced to touch the points, the resulting shape feels more "honest" to the original design.

The Results: Smoothness in Action

The paper tests this method on several models:

The Torus (Donut): Even though the math doesn't create a perfect mathematical donut, the light reflections (isophotes) flow smoothly across the surface, showing no jagged seams.
The Trebol (Three-leaf clover): This model has corners with 3, 4, 5, and 6 sides. The test shows that the surface connects these different shapes seamlessly. The only minor flaw noted is that the edges can look a tiny bit "flat," but the overall flow is continuous.
The Icosahedron (20-sided die): Starting with a shape made of triangles, the method converts it into a smooth surface. The curvature maps show that the surface behaves well, with only small ripples near the edges, which is expected.

Summary

In short, this paper offers a new recipe for turning a rough, multi-sided 3D wireframe into a smooth, continuous skin. It solves the problem of "weird corners" by using special mathematical curves that act like flexible, weighted bridges, ensuring the final surface touches every point of the original design perfectly and flows smoothly from one shape to another.

The author notes that currently, this only works on closed shapes (like a ball), and future work could try to apply it to open shapes (like a bowl) or add even more control over the surface's shape.

Technical Summary: Polyhedral Design with Blended n-Sided Interpolants

Problem Statement

The paper addresses the challenge of generating smooth parametric surfaces that interpolate the vertices of a closed mesh with arbitrary topology. While existing methods based on control polyhedra (such as recursive subdivision, generalized splines, and manifold approaches) often produce approximative surfaces, this work focuses on interpolating the input points directly. The primary difficulty lies in handling irregular vertices (those with valency other than 4), which create non-quadrilateral control nets. Standard tensor-product patches are insufficient for these multi-sided regions, necessitating a specialized parameterization and blending technique to ensure smooth continuity ( $C^k$ ) across the entire surface.

Methodology

1. General Framework

The proposed method constructs a surface by blending local, multi-sided quadratic interpolants. The input is assumed to be a closed mesh; if the mesh contains non-quadrilateral faces, a central split (analogous to a Catmull–Clark subdivision step) is performed to retain original vertices while converting all faces to quads. For each quad in the mesh, the 1-ring neighborhood of its corners defines a multi-sided quadratic control net. Four local patches are generated to interpolate these control points and are subsequently blended together within the parametric region associated with the quad.

2. Regular Meshes (Valency 4)

For regular meshes where all vertices have a valency of 4, the control nets are quadrilateral. The method employs quadratic tensor-product Bézier patches.

Interpolant Definition: The patch $I(u, v)$ is defined using Bernstein polynomials $B^d_k(t)$ . The control points $P_{ij}$ are derived from the corner points ( $C_i$ ), edge points ( $E_i$ ), and the central point ( $M$ ).
Edge Handling: Edge control points are adjusted to $\hat{E}_i = 2E_i - \frac{1}{2}(C_{i-1} + C_i)$ to ensure the quadratic Bézier curve interpolates the original edge point $E_i$ .
Blending: The final surface $S(u, v)$ for a quad is the weighted sum of four local interpolants, each evaluated in a rotated local parameter space. The blending function $\Phi(u, v)$ is the product of two 1-argument Hermite blends ( $k=2$ ), ensuring $C^2$ continuity at the boundaries.

3. Irregular Vertices (Valency $n \neq 4$ )

For irregular vertices, the control net is an $n$ -sided polygon. The paper introduces Quadratic Generalized Bézier (QGB) patches.

QGB Construction: The surface is defined over a regular $n$ -sided polygon using Wachspress coordinates ( $\lambda_i$ ). The interpolant combines contributions from the boundary edges (using quadratic Bézier curves) and a central control point $P_0$ .
Central Point Calculation: $P_0$ is calculated to ensure the patch interpolates the central point $M$ . For $n=4$ , this formulation reduces to the tensor-product case; for $n=3$ , it corresponds to a quadratic triangular Bézier patch.
Parameterization Challenge: A naive bilinear mapping from the quad domain $[0, 1]^2$ to the polygon domain results in discontinuities in adjacent parts of the same interpolant.
Rational Curve Mapping: To achieve $C^\infty$ $C^{\infty}$ continuity (except at the singular origin), the paper proposes a mapping based on the intersection of two pencils of quadratic rational Bézier curves.
- $u$ -curves and $v$ -curves are defined with specific control points and rational weights.
- A non-linear exponential term in the middle control point weight is introduced to handle concave domains (specifically for triangular cases), effectively "pulling back" the curve near the origin.
- The intersection of these curves is computed via nested golden section searches, a process suggested to be precomputed for efficiency.

4. Triangular Subpatches

For the specific case of $n=3$ , the standard QGB patch lacks a central control point. To interpolate the central point $M$ , the method degree-elevates the boundaries to cubic and utilizes a cubic Bézier triangle. This introduces an extra degree of freedom ( $P_{111}$ ) which is solved to satisfy the interpolation constraint at $M$ .

5. Alternative Representations

The paper notes that other multi-sided representations, such as Midpoint and Midpoint Coons patches, are viable. However, experimental results indicated little visual difference in the final models, while the proposed QGB patches proved computationally simpler and more efficient.

Key Contributions

Interpolating Parametric Surfaces: A shift from approximative subdivision schemes to a construction that strictly interpolates the input mesh vertices.
Quadratic Generalized Bézier (QGB) Patches: An extension of Generalized Bézier patches to the quadratic case, allowing for the interpolation of multi-sided control nets.
Rational Curve Parameterization: A novel $C^\infty$ -continuous mapping technique using intersecting rational Bézier curves to handle the parameterization of irregular (non-quadrilateral) domains, solving the discontinuity issues of simpler mappings.
Unified Handling of Topology: A framework that seamlessly blends regular quad patches with multi-sided QGB patches to handle arbitrary mesh topologies.

Results

The method was tested on several models:

Torus (Regular Quadmesh): Demonstrated smooth flow of isophote lines, though the global shape deviated from a perfect torus due to the interpolation constraints.
Trebol Model: A complex object with vertices of valency 3, 4, 5, and 6. Isophote lines confirmed continuous connection and good quality of multi-sided patches. Mean curvature maps showed slight flatness along edges.
Icosahedron: Converted to a quadmesh via Catmull–Clark subdivision. Curvature analysis showed larger values near 5-valent vertices but otherwise small fluctuations corresponding to mesh edges.
Bob Model: A complex model demonstrating the method's capability with environment mapping, contouring, and mean curvature visualization.

Significance and Future Work

The paper claims significance in providing a new, interpolating surface representation that maintains smoothness across arbitrary topologies without relying on subdivision limits. The authors are modest in their claims, noting that while the method works well for closed meshes, several avenues remain for future research:

Generalization to meshes with boundaries.
Incorporation of continuous boundary constraints (positional and cross-derivative interpolation).
Extension with shape parameters.
Improvement of surface quality through normal vector interpolation.

The work is supported by the Hungarian Scientific Research Fund (OTKA, No. 145970).