Rigorous validation and generalization of the adiabatic isometric mapping (AIM) algorithm for embedding two-surface metrics in Euclidean 3-space

The adiabatic isometric mapping (AIM) algorithm of Ray, Miller, Alsing and Yau constructively embeds a polyhedral two-metric — a triangulated surface with prescribed edge lengths — into $\mathbb{R}^3$, by uniformizing the metric with discrete Ricci flow, embedding the constant-curvature target on a sphere with a convexity routine, and pulling the embedding back to the original metric along a conformal path in small, Newton-minimized steps. It empirically realizes Alexandrov-type convex embeddings within edge-length tolerance $10^{-6}$, scales sub-cubically in the vertex number, converges at first order in integrated mean curvature on smooth test surfaces, and embeds a surface just outside a near-extremal Kerr ergosphere that has small negative-curvature regions — but none of these properties is proven, the meshes must be produced externally, and the negative-curvature regime was barely probed. The open problem posed in the paper's conclusion is to complete the validation and generalization of AIM: rigorous upper bounds on the deviation in extrinsic curvature between the essentially equivalent embeddings AIM produces (and their closeness to the unique Alexandrov polyhedron as resolution increases); a general, robust triangulation algorithm producing distmesh-quality meshes from an arbitrary compact two-metric; systematic convergence and robustness analysis for metrics with larger or more complex negative Gaussian curvature and for pathological triangulations; and integration into numerical-relativity pipelines for quasi-local energy computation.

Background

An isometric embedding of a two-dimensional Riemannian metric into Euclidean three-space $\mathbb{R}^3$ is a placement of the surface in space that preserves all intrinsic distances. For smooth metrics on the sphere with strictly positive Gaussian curvature (the intrinsic curvature of the surface, positive on the boundary of a convex body), existence and uniqueness of a convex isometric embedding is Weyl's problem, solved classically by Nirenberg. In the polyhedral category the input is a triangulated surface with prescribed edge lengths (a polyhedral two-metric); the curvature at a vertex is the angle defect, $2\pi$ minus the sum of the incident triangle angles. Alexandrov's theorem states that a polyhedral metric homeomorphic to the sphere with non-negative curvature at every vertex is realized by a unique convex polyhedron in $\mathbb{R}^3$; Bobenko and Izmestiev gave a constructive variational proof of it, and Kane, Price and Demaine gave a pseudopolynomial-time algorithm realizing it, but no polynomial-time constructive algorithm is known.

The computational demand comes largely from general relativity. Quasi-local energy formulas such as the Brown-York and Wang-Yau energies assign an energy to a closed two-surface in a spacetime: the Brown-York construction requires isometrically embedding the surface's induced two-metric in $\mathbb{R}^3$, and the Wang-Yau construction embeds it in Minkowski space (flat spacetime). Surfaces of interest include horizons and ergospheres (the region around a rotating Kerr black hole inside which no observer can remain at rest), whose two-metrics are available only as numerical-relativity grid data (numerical relativity being the numerical solution of Einstein's equations on supercomputers). Existing numerical embedding methods each carry substantive limitations: the spectral approach of Bondarescu, Alcubierre and Seidel relies on symmetry and curvature-sign assumptions; earlier 'wire-mesh' spring models constrain only edge lengths and cannot distinguish extrinsically different realizations of the same intrinsic data; the embedding flow of Jasiulek and Korzyński (which reaches the target embedding as the endpoint of a deformation starting from the round sphere) assumes positive scalar curvature; and the companion algorithm of Tichy, McDonald and Miller, like the others, came without convergence guarantees.

The adiabatic isometric mapping (AIM) algorithm of Ray, Miller, Alsing and Yau (2015) works directly on a polyhedral two-metric in three stages. First, discrete Ricci flow — a combinatorial analogue of Hamilton's Ricci flow that evolves vertex-wise conformal (scale) factors so as to spread curvature uniformly — uniformizes the metric to constant positive curvature. Second, the uniformized surface is embedded on a round sphere by local triangle patches, guided by annealing (Newton minimization of edge-length residuals) and a convexity routine, a heuristic that detects vertices where the incident triangles fold inward and reflects them outward toward convex solutions. Third, a guided adiabatic pull-back (GAP) deforms this embedding back to the original metric along a linear path of conformal factors in small time steps, each step Newton-minimized with extrapolated initial guesses; the small steps are what keep the iteration in the basin of a smooth global minimum instead of a crumpled one. Empirically, AIM recovers edge lengths to tolerances of $10^{-6}$, scales sub-cubically in the number of vertices (dominated by the Ricci-flow stage), converges at first order in the integrated mean curvature (the surface integral of the mean curvature, i.e. the average of the two principal curvatures, computed on triangulations by the dihedral-angle-based discretization of Conboye, Miller and Ray) for an ellipsoid and a drum-like test surface, and produced a smooth embedding of a surface just outside the ergosphere of a near-extremal Kerr black hole ($a/M=0.975$, where $a/M$ is the black hole's dimensionless spin parameter, with negative Gaussian curvature near the poles) with $0.62\%$ error in integrated mean curvature. The convexity routine is only expected to work for 'well-posed' triangulations in which every triangle angle lies between $\pi/6$ and $2\pi/3$, and the meshes themselves must be supplied externally: distmesh, the standard MATLAB mesh generator of Persson and Strang, requires an implicit description of the surface in space and cannot mesh an abstract two-metric. None of AIM's accuracy, robustness, or applicability properties is proven; establishing them is the open problem.

Problem Statement

Complete the validation and generalization program that Ray, Miller, Alsing and Yau pose in the conclusion of their paper on the adiabatic isometric mapping (AIM) algorithm, so that AIM becomes a mathematically validated and generally applicable method for isometrically embedding compact two-metrics (metrics on compact two-dimensional surfaces specified intrinsically, for example by the edge lengths of a triangulation) in $\mathbb{R}^3$. The program has four subgoals.

(1) Extrinsic-curvature error bounds. Provide a rigorous upper bound on the deviation in extrinsic curvature — the curvature of the surface as it sits in $\mathbb{R}^3$, determined by the embedding — between the essentially equivalent approximate embeddings (the several distinct embeddings AIM produces for the same polyhedral metric at a given resolution, each realizing the prescribed edge lengths to within the Newton tolerance), and quantify how these embeddings and their extrinsic-curvature observables behave as resolution increases — in particular their closeness to the unique Alexandrov polyhedron (the unique convex polyhedron realizing a convex polyhedral metric, whose existence and uniqueness is Alexandrov's theorem) for convex polyhedral metrics.

(2) General robust triangulation. Develop an algorithm, comparable in output quality to distmesh (the standard MATLAB mesh generator of Persson and Strang, which requires an implicit description of the surface in ambient space), that produces well-posed triangulations — with triangle angles in a range such as $(\pi/6, 2\pi/3)$ and consistent quality as resolution increases — given an arbitrary compact two-metric (intrinsic data only, no ambient or implicit surface description), with edge lengths realizing the metric to the accuracy AIM requires.

(3) Systematic convergence and robustness analysis in the negative-curvature regime. Characterize AIM's behavior and the convergence of its extrinsic-curvature observables on polyhedral metrics with larger or more complex regions of negative Gaussian curvature than the single perturbed Kerr-ergosphere test reported (a surface just outside the ergosphere of a near-extremal rotating black hole, carrying small negative-curvature regions), including degenerate or pathological triangulations on which the convexity routine (the heuristic step that detects inward-folding vertices and reflects them outward toward convex solutions) may fail, establishing — or refuting — convergence and correctness in these regimes.

(4) Application to quasi-local energy in numerical relativity. Integrate AIM into a numerical-relativity workflow: interpolate grid metric data onto intrinsic triangulations, and validate AIM-based computation of quasi-local energies (such as the Brown-York energy, whose evaluation requires isometrically embedding the surface's induced two-metric in $\mathbb{R}^3$) on realistic simulation data against analytic benchmarks such as Kerr surfaces.

These subgoals form one program: the paper itself identifies the absence of consistent high-quality triangulations (subgoal 2) as the main obstacle to the missing convergence analysis of the ergosphere test (subgoal 3); the error bounds (subgoal 1) quantify the accuracy that the quasi-local energy application (subgoal 4) needs; and subgoal 3 extends the tested scope to the surfaces numerical relativity actually produces. Performance engineering of the implementation (preconditioning of the Newton routines, parallelization) is instrumental to this program, since the source paper itself states that optimizing AIM 'was not our goal or intention'.

References

1. S. Ray, W. A. Miller, P. M. Alsing, S.-T. Yau, 'Adiabatic isometric mapping algorithm for embedding 2-surfaces in Euclidean 3-space', Classical and Quantum Gravity 32 (2015) 235012. DOI: 10.1088/0264-9381/32/23/235012; https://arxiv.org/abs/1503.07556
2. G. Gao, K. Hu, B. Li, G. Zhang, 'Finite element methods for isometric embedding of Riemannian manifolds', arXiv:2602.18722 (2026); https://arxiv.org/abs/2602.18722
3. W. A. Miller, S. Ray, M.-T. Wang, S.-T. Yau, 'Wang and Yau's quasi-local energy for an extreme Kerr spacetime', Classical and Quantum Gravity 35 (2018) 055007. DOI: 10.1088/1361-6382/aaa625; https://arxiv.org/abs/1708.07532
4. D. Pook-Kolb, B. Zhao, L. Andersson, B. Krishnan, S.-T. Yau, 'Properties of quasilocal mass in binary black hole mergers', Physical Review D 108 (2023) 124031. DOI: 10.1103/PhysRevD.108.124031; https://arxiv.org/abs/2308.10906
5. V. A. Alexandrov, 'Recognition of Affine-Equivalent Polyhedra by Their Natural Developments', Siberian Mathematical Journal 64 (2023). DOI: 10.1134/S0037446623020027; https://arxiv.org/abs/2106.13659
6. L. Chantry, V. Cayatte, C. Sauty, 'Conformal representation of Kerr space-time poloidal sub-manifolds', Classical and Quantum Gravity 37 (2020). DOI: 10.1088/1361-6382/ab8399; https://doi.org/10.1088/1361-6382/ab8399
7. T. Gabriel, J.-F. Remacle, C. Geuzaine, 'Intrinsic Meshing of Closed Surfaces Using Geodesic Distances', arXiv:2607.04989 (2026); https://arxiv.org/abs/2607.04989
8. W. Tichy, J. R. McDonald, W. A. Miller, 'New efficient algorithm for the isometric embedding of 2-surface metrics in three dimensional Euclidean space', Classical and Quantum Gravity 32 (2015) 015002. DOI: 10.1088/0264-9381/32/1/015002; https://doi.org/10.1088/0264-9381/32/1/015002
9. M. Jasiulek, M. Korzyński, 'Isometric embeddings of 2-spheres by embedding flow for applications in numerical relativity', Classical and Quantum Gravity 29 (2012) 155010. DOI: 10.1088/0264-9381/29/15/155010; https://doi.org/10.1088/0264-9381/29/15/155010
10. M. Bondarescu, M. Alcubierre, E. Seidel, 'Isometric embeddings of black-hole horizons in three-dimensional flat space', Classical and Quantum Gravity 19 (2002) 375. DOI: 10.1088/0264-9381/19/2/311; https://doi.org/10.1088/0264-9381/19/2/311
11. D. Kane, G. N. Price, E. D. Demaine, 'A Pseudopolynomial Algorithm for Alexandrov's Theorem', in Algorithms and Data Structures (WADS 2009), Lecture Notes in Computer Science 5664, Springer (2009) 435-446. DOI: 10.1007/978-3-642-03367-4_38; https://doi.org/10.1007/978-3-642-03367-4_38
12. A. I. Bobenko, I. Izmestiev, 'Alexandrov's theorem, weighted Delaunay triangulations, and mixed volumes', Annales de l'Institut Fourier 58 (2008) 447-505. DOI: 10.5802/aif.2358; https://doi.org/10.5802/aif.2358
13. L. Nirenberg, 'The Weyl and Minkowski problems in differential geometry in the large', Communications on Pure and Applied Mathematics 6 (1953) 337-394. DOI: 10.1002/cpa.3160060303; https://doi.org/10.1002/cpa.3160060303
14. J. D. Brown, J. W. York, Jr., 'Quasilocal energy and conserved charges derived from the gravitational action', Physical Review D 47 (1993) 1407-1419. DOI: 10.1103/PhysRevD.47.1407; https://doi.org/10.1103/PhysRevD.47.1407
15. M.-T. Wang, S.-T. Yau, 'Quasilocal mass in general relativity', Physical Review Letters 102 (2009) 021101. DOI: 10.1103/PhysRevLett.102.021101; https://doi.org/10.1103/PhysRevLett.102.021101
16. R. Conboye, W. A. Miller, S. Ray, 'Distributed mean curvature on a discrete manifold for Regge calculus', Classical and Quantum Gravity 32 (2015) 185009. DOI: 10.1088/0264-9381/32/18/185009; https://doi.org/10.1088/0264-9381/32/18/185009
17. P.-O. Persson, G. Strang, 'A Simple Mesh Generator in MATLAB', SIAM Review 46 (2004) 329-345. DOI: 10.1137/S0036144503429121; https://doi.org/10.1137/S0036144503429121
18. J.-D. Boissonnat, R. Dyer, A. Ghosh, 'Delaunay Triangulation of Manifolds', Foundations of Computational Mathematics 18 (2018) 399-431. DOI: 10.1007/s10208-017-9344-1; https://doi.org/10.1007/s10208-017-9344-1