Manifold splines without affine structure: singularity-free spline schemes for high-genus surfaces and the singularity-count versus distortion trade-off

Classical manifold splines extend planar spline theory to surfaces of arbitrary topology by covering the domain with charts whose transition functions are affine, so that polar-form (blossoming) spline machinery transfers verbatim; an affine atlas, however, cannot cover a closed surface of genus 2 or more without singular points, by the Euler-class/Gauss-Bonnet obstruction. Gu, He, Jin, Luo, Qin and Yau (2008) pushed this framework to its topological limit, using discrete Ricci flow to concentrate all Gaussian curvature at a single extraordinary point - the proven lower bound for closed high-genus surfaces - and stated two open directions: (i) spline schemes that do not depend on an affine atlas, ideally singularity-free on high-genus surfaces, and (ii) a systematic account of the trade-off between the number of singularities and the area distortion of the parameterization, which their construction leaves to ad hoc user choice.

Background

Splines - piecewise polynomial or rational functions glued together with prescribed smoothness - are the workhorse representation for free-form shapes in computer-aided geometric design (CAGD). Classical bivariate splines are defined over planar parameter domains (triangulations or tensor-product grids), so a closed surface with handles must be represented by trimming and stitching many patches, with continuity loss at the seams. Manifold splines, in the sense of Grimm and Hughes (1995) and Gu, He and Qin (2006), remove the planar restriction by taking the surface itself as the parameter domain: the surface is covered by coordinate charts and spline functions are assembled chart by chart. The gluing is harmless only when chart transitions preserve the function class. If every overlap of two charts is related by an affine transition $x \mapsto Ax + b$, with $A$ an invertible $2 \times 2$ matrix - an affine atlas - then polynomials map to polynomials and the entire planar theory transfers: polar forms (the multi-affine 'blossoming' characterization of B-spline segments), local bases with partition of unity, and degree elevation. Gu, He and Qin (2006) proved that a surface admits polar-form manifold splines if and only if it admits an affine atlas, and identified the obstruction as the Euler class. Concretely, an affine atlas amounts to a flat metric away from cone singularities - points where Gaussian curvature is concentrated and the parameterization degenerates - and by the Gauss-Bonnet theorem a closed, oriented surface of genus $g$ (the number of handles; a surface with two handles has $g = 2$) must carry total curvature $2\pi(2 - 2g)$, which for $g \neq 1$ forces at least one cone point. Equivalently, a classical theorem of Milnor (1958) rules out a flat connection on the tangent bundle of any closed surface except the torus. Prior atlas constructions built from holomorphic 1-forms (complex-analytic differential forms used to flatten a surface conformally) carried $|2g - 2|$ singular points on a genus-$g$ surface.

Gu, He, Jin, Luo, Qin and Yau (2008) pushed the framework to its topological limit. Using discrete Ricci flow - a combinatorial analogue, on circle-packing metrics (edge lengths derived from prescribed radii of mutually tangent circles), of the curvature-driven metric evolution from differential geometry - they conformally deformed the domain metric so that all Gaussian curvature concentrates at a single user-chosen vertex, producing an affine atlas with exactly one extraordinary point, the theoretical lower bound for closed surfaces of genus $\geq 2$. Over this atlas they built manifold triangular B-splines of degree 3 that are $C^2$ away from the singular point, filling the hole left by its one-ring neighborhood with planar patches joined with $G^1$ continuity (geometric continuity of order 1 means a shared tangent plane after reparameterization; $G^r$ requires matching derivatives up to order $r$ after some reparameterization). Constructions without an affine atlas already existed: Ying and Zorin (2004) built $C^\infty$ surfaces of arbitrary topology from manifold charts blended by non-polynomial weight functions.

The 2008 paper's own discussion and conclusion state two limitations that define the present problem. First, the framework is intrinsically tied to affine atlases, which closed high-genus surfaces cannot carry without a singular point; hence 'the most natural spline solutions for high genus manifolds should not depend on their affine structures', and future work should investigate 'spline schemes which are not based on the affine structure of the underlying surface domain' - ideally singularity-free. Second, concentrating all curvature at one point inflates the area distortion of the flattening (the nonuniformity of area scaling between the surface and the parameter domain), while spreading curvature over more singularities lowers distortion at the price of many more extraordinary points; the same authors' polycube splines (Wang, He, Li, Gu and Qin, 2008) - splines built over polycube-shaped domains, unions of axis-aligned boxes - sit at the many-singularities/low-distortion end. The paper leaves the balance to ad hoc user choice and offers no systematic treatment of the trade-off.

Problem Statement

The problem posed by Gu, He, Jin, Luo, Qin and Yau (2008) at the conclusion of their single-extraordinary-point manifold spline framework is two-fold, with one overarching objective: a spline theory for closed high-genus surfaces that is free of the affine-atlas bottleneck. (1) Develop spline schemes for surfaces of arbitrary topology - in particular closed surfaces of genus $\geq 2$ - that do not depend on an affine atlas of the domain, with the aim of singularity-free spline surfaces on high-genus manifolds (for a scheme that does not use an affine atlas, a singular point is any point at which the parameterization or the spline construction degenerates - for example a cone point of the parameterization metric, or an extraordinary point at which the charts, transition maps, or local spline basis fail the claimed regularity; singularity-free means that no such point exists, so relocating or renaming a singular point does not remove it), while retaining as much of the classical spline apparatus as possible (a genuine function space with local support, partition of unity, controllable smoothness, and usable bases). (2) Replace the ad hoc user choice between few singularities (high area distortion) and many singularities (low distortion) by a systematic account of the trade-off between the number of singular points and the quality - notably area distortion - of the resulting parameterization.

Part (1) has several published realizations - hyperbolic-domain (Zeng, He, Xia, Gu and Qin, 2009), projective-linear (Peters and Fan, 2010), rational geometric (Beccari, Gonsor and Neamtu, 2014), homogeneous (Beccari and Neamtu, 2016), and orbifold (Beccari and Prautzsch, 2022) spline schemes - so a new contribution under (1) must demonstrably go beyond these published realizations. What remains without a dedicated treatment is (2) - an account of the trade-off between the number of singular points and the quality - notably area distortion - of the resulting parameterization, demonstrated to control that trade-off systematically rather than by heuristic choice - together with any genuinely new class of schemes under (1). The two parts are independent components of the objective.

References

1. Xianfeng Gu, Ying He, Miao Jin, Feng Luo, Hong Qin, Shing-Tung Yau. Manifold splines with a single extraordinary point. Computer-Aided Design 40(6):676-690, 2008. DOI: 10.1016/j.cad.2008.01.008. https://doi.org/10.1016/j.cad.2008.01.008
2. Xianfeng Gu, Ying He, Hong Qin. Manifold splines. Graphical Models 68(3):237-254, 2006. DOI: 10.1016/j.gmod.2006.03.004. https://doi.org/10.1016/j.gmod.2006.03.004
3. Lexing Ying, Denis Zorin. A simple manifold-based construction of surfaces of arbitrary smoothness. ACM Transactions on Graphics 23(3):271-275, 2004. DOI: 10.1145/1015706.1015714. https://doi.org/10.1145/1015706.1015714
4. Cindy M. Grimm, John F. Hughes. Modeling surfaces of arbitrary topology using manifolds. Proceedings of SIGGRAPH '95, pp. 359-368, 1995. DOI: 10.1145/218380.218475. https://doi.org/10.1145/218380.218475
5. Wei Zeng, Ying He, Jiazhi Xia, Xianfeng Gu, Hong Qin. C∞ smooth freeform surfaces over hyperbolic domains. Proceedings of the 2009 SIAM/ACM Joint Conference on Geometric and Physical Modeling (SPM 2009), pp. 367-372, 2009. DOI: 10.1145/1629255.1629305. https://doi.org/10.1145/1629255.1629305
6. Jörg Peters, Jianhua Fan. The projective linear transition map for constructing smooth surfaces. 2010 Shape Modeling International Conference (SMI 2010), pp. 124-130, 2010. DOI: 10.1109/smi.2010.26. https://doi.org/10.1109/smi.2010.26
7. Hongyu Wang, Ying He, Xin Li, Xianfeng Gu, Hong Qin. Polycube splines. Computer-Aided Design 40(6):721-733, 2008. DOI: 10.1016/j.cad.2008.01.012. https://doi.org/10.1016/j.cad.2008.01.012
8. Carolina Vittoria Beccari, Daniel E. Gonsor, Marian Neamtu. RAGS: Rational geometric splines for surfaces of arbitrary topology. Computer Aided Geometric Design 31(2):97-110, 2014. DOI: 10.1016/j.cagd.2013.11.004. https://doi.org/10.1016/j.cagd.2013.11.004
9. Carolina Vittoria Beccari, Marian Neamtu. On constructing RAGS via homogeneous splines. Computer Aided Geometric Design 43:109-122, 2016. DOI: 10.1016/j.cagd.2016.02.010. https://doi.org/10.1016/j.cagd.2016.02.010
10. Bernard Mourrain, Raimundas Vidunas, Nelly Villamizar. Dimension and bases for geometrically continuous splines on surfaces of arbitrary topology. Computer Aided Geometric Design 45:108-133, 2016. DOI: 10.1016/j.cagd.2016.03.003 (arXiv:1509.03274). https://doi.org/10.1016/j.cagd.2016.03.003
11. Carolina Vittoria Beccari, Hartmut Prautzsch. Quadrilateral orbifold splines. In Geometric Challenges in Isogeometric Analysis, Springer INdAM Series, pp. 1-18, 2022. DOI: 10.1007/978-3-030-92313-6_1. https://doi.org/10.1007/978-3-030-92313-6_1
12. Angelos Mantzaflaris, Bernard Mourrain, Nelly Villamizar, Beihui Yuan. An algebraic framework for geometrically continuous splines. Mathematics of Computation 95(358):925-968, 2025. DOI: 10.1090/mcom/4068 (arXiv:2305.09096). https://doi.org/10.1090/mcom/4068
13. Ashish Myles, Denis Zorin. Global parametrization by incremental flattening. ACM Transactions on Graphics 31(4):1-11, 2012. DOI: 10.1145/2185520.2185605. https://doi.org/10.1145/2185520.2185605
14. Ashish Myles, Denis Zorin. Controlled-distortion constrained global parametrization. ACM Transactions on Graphics 32(4):1-14, 2013. DOI: 10.1145/2461912.2461970. https://doi.org/10.1145/2461912.2461970
15. Marcel Campen, Denis Zorin. Similarity maps and field-guided T-splines. ACM Transactions on Graphics 36(4):1-16, 2017. DOI: 10.1145/3072959.3073647. https://doi.org/10.1145/3072959.3073647
16. Xiaopeng Zheng, Yiming Zhu, Wei Chen, Na Lei, Zhongxuan Luo, Xianfeng Gu. Quadrilateral mesh generation III: Optimizing singularity configuration based on Abel-Jacobi theory. Computer Methods in Applied Mechanics and Engineering 387:114146, 2021. DOI: 10.1016/j.cma.2021.114146. https://doi.org/10.1016/j.cma.2021.114146
17. Yuan-guang Wang. High genus surface parameterization using the Euclidean Ricci flow method. Scientific Reports 15, article 17784, 2025. DOI: 10.1038/s41598-025-97421-5. https://doi.org/10.1038/s41598-025-97421-5
18. John Milnor. On the existence of a connection with curvature zero. Commentarii Mathematici Helvetici 32(1):215-223, 1958. DOI: 10.1007/bf02564579. https://doi.org/10.1007/bf02564579