Random Lie Brackets that Induce Torsion: A Model for Noisy Vector Fields

Didong Li, Sayan MukherjeearXiv 2020 · DOI · arXiv

Abstract

We define and study a random Lie bracket that induces torsion in expectation. Almost all stochastic analysis on manifolds have assumed parallel transport. Mathematically this assumption is very reasonable. However, in many applied geometry and graphics problems parallel transport is not achieved, the "change in coordinates" are not exact due to noise. We formulate a stochastic model on a manifold for which parallel transport does not hold and analyze the consequences of this model with respect to classic quantities studied in Riemannian geometry. We first define a stochastic lie bracket that induces a stochastic covariant derivative. We then study the connection implied by the stochastic covariant derivative and note that the stochastic lie bracket induces torsion. We then state the induced stochastic geodesic equations and a stochastic differential equation for parallel transport. We also derive the curvature tensors for our construction and a stochastic Laplace-Beltrami operator. We close with a discussion of the motivation and relevance of our construction.

Introduction

In classical Riemannian geometry and most stochastic manifold analysis, researchers assume the "torsion-free" condition, specifically the Levi-Civita connection. This assumption implies that parallel transport is exact: if a piece of data is moved along a path on a surface, the transformation of its coordinates is perfectly determined by the surface's intrinsic shape.

However, in practical applications such as computer graphics or biological morphometrics (e.g., measuring the shapes of bones or teeth), this assumption often fails. Noise in data collection or errors in mapping one shape to another mean that coordinate changes are not exact. Mathematically, this "noise" manifests as torsion. This paper introduces a robust stochastic model that incorporates torsion as a fundamental part of the manifold's geometry rather than treating it as an external error term.

Methodology: Randomization of Vector Fields

The framework is built through a systematic randomization of vector fields:

  1. Random Vector Fields: A randomized vector field $\tilde{X}$ is defined as the product of a standard smooth vector field $X$ and a random smooth function $\epsilon$ (where $\mathbb{E}[\epsilon] = 1$).
  2. Stochastic Connection: A new "stochastic covariant derivative" $\tilde{D}_X Y$ is defined based on these randomized fields. Unlike standard derivatives, this one does not assume the Lie bracket (describing how vector fields commute) is deterministic.
  3. Geometric Re-derivation: Using this new connection, fundamental Riemannian structures are re-derived:
  4. Christoffel Symbols: Coordinate building blocks now include terms related to the gradient of the noise $\epsilon$.
  5. Geodesic Equations: Equations for "shortest paths" are derived as stochastic differential equations.
  6. Curvature Tensors: The Riemann, Ricci, and Scalar curvatures are calculated under the influence of noise.
  7. Laplace-Beltrami Operator: A stochastic version of this operator is derived, which is critical for heat diffusion and manifold learning.

Motivating Example: Biological Morphometrics

While primarily theoretical, the model is motivated by real-world data challenges. The authors utilize 3D scans of lemur teeth as a qualitative example. When attempting to map points from one tooth to another, the "noise" in the correspondence map reflects the exact phenomenon this torsion-based model aims to describe.

Mapping landmarks between 3D scans of lemur teeth. The noise in these correspondence maps motivates a geometric model that incorporates torsion.

Key Insights and Results

Torsion as Geometric Noise

The most profound insight is that torsion is the natural geometric language for "noisy" parallel transport. The paper demonstrates that while the Sectional Curvature remains unchanged (as randomization only rescales vectors within the same plane), the Ricci and Scalar Curvatures scale by $\epsilon^3$.

The Topology Obstruction

The paper demonstrates that the classical Chern-Gauss-Bonnet theorem—which links a surface's local curvature to its global topology—does not hold in this stochastic setting. The local addition of noise creates an "obstruction" that prevents the simple recovery of global topological properties.

Modified Diffusion

The stochastic Laplace-Beltrami operator includes an extra term: $\tilde{\Delta} f = \epsilon^2 \Delta f + 2\epsilon \nabla f(\epsilon)$. This suggests that information diffuses differently on a noisy manifold than on a smooth one, with direct implications for algorithms like Diffusion Maps.

Existence of Geodesics

The authors prove that if the noise function $\epsilon$ is sufficiently smooth ($C^2$), unique "geodesics in expectation" exist, allowing for a consistent way to define paths across noisy geometries. Furthermore, a version of the classical Divergence Theorem still holds in this noisy regime, which is vital for physical simulations.

Limitations and Future Work

The current model requires the noise $\epsilon$ to be at least $C^2$ smooth, excluding "rougher" noise types like pure Brownian motion. Additionally, while the theory is established, numerical libraries for implementing these stochastic operators on discrete structures like triangle meshes are yet to be developed. Future research may extend this framework to more general structures like Finsler geometries or Ehresmann connections to model even more complex types of data uncertainty.

Summary updated

Abstract

We define and study a random Lie bracket that induces torsion in expectation. Almost all stochastic analysis on manifolds have assumed parallel transport. Mathematically this assumption is very reasonable. However, in many applied geometry and graphics problems parallel transport is not achieved, the "change in coordinates" are not exact due to noise. We formulate a stochastic model on a manifold for which parallel transport does not hold and analyze the consequences of this model with respect to classic quantities studied in Riemannian geometry. We first define a stochastic lie bracket that induces a stochastic covariant derivative. We then study the connection implied by the stochastic covariant derivative and note that the stochastic lie bracket induces torsion. We then state the induced stochastic geodesic equations and a stochastic differential equation for parallel transport. We also derive the curvature tensors for our construction and a stochastic Laplace-Beltrami operator. We close with a discussion of the motivation and relevance of our construction.
BibTeX
@misc{li2020random,
  author = {Didong Li and Sayan Mukherjee},
  title = {{Random Lie Brackets that Induce Torsion: A Model for Noisy Vector Fields}},
  year = {2020},
  eprint = {2007.07309},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2007.07309},
}
Figure