Extrinsic Vector Field Processing

Abstract: We propose a novel discretization of tangent vector fields for triangle meshes. Starting with a Phong map continuously assigning normals to all points on the mesh, we define an extrinsic bases for continuous tangent vector fields by using the Rodrigues rotation to transport tangent vectors assigned to vertices to tangent vectors in the interiors of the triangles. As our vector fields are continuous and weakly differentiable, we can use them to define a covariant derivative field that is evaluatable almost-everywhere on the mesh. Decomposing the covariant derivative in terms of diagonal multiple of the identity, anti-symmetric, and trace-less symmetric components, we can define the standard operators used for vector field processing including the Hodge Laplacian energy, Connection Laplacian energy, and Killing energy. Additionally, the ability to perform point-wise evaluation of the covariant derivative also makes it possible for us to define the Lie bracket.

• The paper presents a novel finite element discretization for tangent vector fields on triangle meshes using extrinsically defined bases.
• It introduces a differentiable framework for computing covariant derivatives and Lie brackets pointwise, enhancing geometric processing.
• Experimental evaluations demonstrate competitive spectral accuracy and robust performance on both regular and irregular mesh tessellations.

Extrinsic Vector Field Processing: A Technical Summary

Introduction and Motivation

The paper "Extrinsic Vector Field Processing" (2601.10621) introduces a novel finite element discretization for tangent vector fields defined over triangle meshes embedded in $\mathbb{R}^3$. The key innovation is the use of extrinsically defined, continuous, and weakly differentiable tangent vector field bases constructed from vertex normals via Phong interpolation and parallel transport using the Rodrigues rotation. This framework addresses limitations in previous vector field discretization approaches, particularly regarding smoothness and the ability to evaluate covariant derivatives and Lie brackets pointwise over the mesh.

Methodology

Vector Field Basis Construction

This approach begins with a triangle mesh equipped with per-vertex normals, which are smoothly interpolated over the domain using Phong interpolation to define a continuous normal field. Given a tangent vector at a vertex, parallel transport is performed via the minimal-angle Rodrigues rotation mapping the vertex normal to the interpolated normal at any interior point of an incident triangle. The vector is then scaled by the corresponding barycentric coordinate to ensure partition-of-unity over each triangle.

Figure 1: The unit-right triangle serves as the reference domain for basis construction and integration.

This construction leads to a basis of continuous tangent vector fields that are everywhere orthogonal to the interpolated normal field, providing an explicit and differentiable formulation suitable for finite element assembly.

Figure 2: Smooth interpolation of sparse vectors on a surface using Dirichlet energies induced by the Connection (center) and Hodge (right) Laplacians, demonstrating basis continuity and vector field smoothness.

Covariant Differentiation and Energies

With the explicit vector-field basis, the covariant derivative is both pointwise evaluatable and decomposable into canonical components—scalar multiple of the identity, anti-symmetric (curl), and trace-free symmetric parts. This supports the direct definition of standard energies in vector field processing:

• Dirichlet/Connection Laplacian Energy: Penalizes deviation from parallelism.
• Hodge Laplacian Energy: Associated with harmonic vector fields and their decomposition.
• Killing Energy: Measures deviation from isometries (infinitesimal rigid motions).

The methodology supports integration over the surface using numerical quadrature, and finite element assembly is formulated for global system construction.

Lie Bracket Computation

Crucially, the explicit differentiable basis enables pointwise evaluation of the Lie bracket of vector fields, a nonlinear operation central to differential geometry:

$$[X, Y](p) = \nabla X|_p \cdot Y(p) - \nabla Y|_p \cdot X(p)$$

The bracket is projected onto the vector-field basis using the mass matrix to obtain a weak least-squares representation.

Evaluation and Comparative Analysis

Geometric Applications

The proposed discretization is evaluated through interpolation and transport tasks (Figure 2, 3), geometry-aware vector diffusion (Figure 3), and spectral analysis (Figure 4).

Figure 3: Vector Heat method application—input constraints (left) and diffused vector field (right)—showing robust parallel transport behavior using the extrinsic basis.

Figure 4: Four smallest eigenvectors for Connection (top), Hodge (middle), and Killing (bottom) energies, illustrating modal structure and harmonic vector field recovery.

Key observations include the non-singularity of the Connection Laplacian energy and the identification of harmonic fields on genus-one surfaces via Hodge energies.

Spectral Accuracy

A spectral analysis on random isotropic and anisotropic triangulations of the unit sphere quantifies the accuracy of the discretized Connection Laplacian (Figure 5). The method achieves error levels indistinguishable from prior intrinsic approaches [Knöppel et al. 2013] and competitive with edge-based and operator-based vector field discretizations.

Figure 5: Mean differences between analytic and estimated connection Laplacian eigenvalues as a function of index, for isotropic (top) and anisotropic (bottom) sphere triangulations.

For Hodge Laplacians, the method recovers spectra consistent with the co-tangent Laplacian at higher frequencies and approximates harmonic vector fields, but marginally lags Whitney basis approaches at the lowest frequencies (Figure 6).

Figure 6: Relative eigenvalue difference between Hodge Laplacian and co-tangent Laplacian; genus $g$ and ratio $\rho$ of low-frequency eigenvalues are provided for context.

Robustness to Mesh Quality

Analysis demonstrates invariance under pointwise tangent plane rotation by $90^\circ$ and insensitivity to mesh aspect ratios, except in degenerate triangulations where numerical error increases (Figure 7).

Figure 7: Distributions of triangle aspect ratios for models used in spectral evaluations, underscoring robustness for high-quality meshes.

Lie Bracket Quantitative Evaluation

Projection-based recovery of Lie brackets is tested against analytic solutions for various tessellations and vector field bandwidths (Figure 8, 10). The method outperforms operator-based approaches [Azencot et al. 2013, Stein et al. 2020] under refinement and on regular meshes, with error increasing for higher frequency fields and lower resolution.

Figure 8: Visualization of Lie bracket estimation for $b=10$ vector fields on regular (top) and irregular (bottom) sphere tessellations. Analytic solution (rightmost) matches closely with the finite element approach on regular meshes.

Figure 9: Relative error versus vector field frequency and resolution, showing improved bracket estimation with mesh refinement.

Discussion and Implications

The extrinsic approach outlined brings several practical and theoretical implications:

• Continuous, Differentiable, and Extrinsic: Unlike piecewise-constant or discontinuous bases, this method yields $C^0$ vector fields compatible with smooth energies and nonlinear tangent-space operations.
• Lie Bracket Evaluation: Direct, finite element-compatible computation of the Lie bracket positions the approach as a bridge between geometry processing and geometric deep learning, particularly in applications requiring expressive, higher-order operators.
• Spectral Processing: Accurate low- and high-frequency eigenmode recovery enables advanced applications in deformation, anisotropic diffusion, and frame field design.
• Comparison to Intrinsic/Whitney/Edge-Based Bases: Though not perfectly mimetic, the proposed method offers a practical trade-off, capturing geometric structure while enabling pointwise and global operations inaccessible to many competing approaches.

Future Prospects

Potential extensions include higher-order elements, frame and tensor field generalizations, and dedicated optimization of per-vertex normals or transport mappings for improved geometric fidelity, particularly in the presence of noisy or imperfect input meshes. The framework is also a natural candidate for integration with neural implicit representations [Park et al. 2019], neural geometry processing [Williamson & Mitra 2025], or in hybrid extrinsic-intrinsic pipelines.

Conclusion

This work presents an extrinsically defined, vector field discretization paradigm for triangle meshes that is both continuous and weakly differentiable, supporting the explicit computation of covariant derivatives, Lie brackets, and a suite of canonical vector energies. Numerical experiments confirm its competitiveness with state-of-the-art intrinsic and edge-based methods for spectral properties, bracket computation, and geometric processing tasks. The explicit basis functions, robust handling of nonlinear operators, and efficient mass and stiffness assembly make it a compelling foundation for both pure geometry processing and future geometric learning systems.