New Exact and Numerical Solutions of the (Convection-)Diffusion Kernels on SE(3)

Jorg M. Portegies, Remco Duits

In the fields of medical imaging and robotics, researchers frequently encounter "elongated structures"—the intricate web of blood vessels in a retina, the complex pathways of nerve fibers in the brain, or the trajectories of a moving agent. To process this data effectively, it is often necessary to "lift" 3D images into higher-dimensional spaces that explicitly include orientation. This transformation allows algorithms to distinguish between crossing structures that would otherwise appear as a single, ambiguous intersection in a standard 3D volume.

A landmark paper by J.M. Portegies and R. Duits, titled "New Exact and Numerical Solutions of the (Convection-)Diffusion Kernels on SE(3)," provides the mathematical "secret sauce" for this process. They have derived the first exact analytical solutions for the kernels that govern how information propagates and connects within the space of 3D positions and orientations.

The Challenge of 3D Orientation

When enhancing or completing "lines" in an image, researchers use diffusion to smooth along a path or convection-diffusion to "predict" where a path is likely to go. While these processes are well-understood in 2D (on the group $SE(2)$), moving to 3D ($SE(3)$) represents a massive jump in mathematical complexity. This is primarily because 3D rotations are non-commutative—the order in which you rotate an object fundamentally changes its final orientation.

Conceptual illustration of path prediction and completion using convection-diffusion processes, where $\gamma(t)$ represents the evolving trajectory.

Historically, researchers working in 3D relied on rough numerical approximations or "local" Gaussian kernels. These methods often failed to perfectly respect the underlying geometry of the space, leading to inaccuracies when tracking brain fibers in diffusion MRI (dMRI) or planning robotic movement. The lack of exact solutions meant there was no "gold standard" for accuracy and no rigorous basis for developing faster, more reliable algorithms.

Methods: A Bridge to Mathematical Physics

The authors tackled this problem by blending harmonic analysis with differential geometry. Their approach involved several key innovations:

  1. Spatial Fourier Transform: They began by applying a Fourier transform to the spatial variables of the governing Partial Differential Equations (PDEs). This step simplified the complex 3D problem into a more manageable second-order differential operator.
  2. Eigenfunction Decomposition: The core breakthrough was demonstrating that the eigenfunctions of this operator are spheroidal wave functions (and their generalized counterparts). These functions are well-known in physics for solving the wave equation in spheroidal coordinates, but their application to $SE(3)$ diffusion is a novel and powerful bridge between fields.
  3. The Resolvent Operator: Beyond just looking at how a process evolves over time, the authors solved the "time-integrated" version of the process, known as the resolvent operator. This is practically essential for tasks like Tikhonov regularization, where one needs to understand the total "influence" of a point over all time.
  4. Symmetry-Preserving Approximations: They developed a new analytic approximation based on the logarithm on $SE(3)$. Crucially, they identified a specific parameterization that ensures the approximation shares the same fundamental symmetries as the exact solution, fixing flaws in previous Gaussian models.

Key Results and Visual Proof

The paper provides explicit series expansions for kernels on the quotient space $\mathbb{R}^3 \rtimes S^2$. Furthermore, the authors extended the "August algorithm"—originally designed for 2D—to 3D, providing a high-speed, matrix-based method to compute these kernels using spherical harmonics.

Glyph field visualizations and fiber reconstructions demonstrating how the kernels effectively flow along orientations to enhance complex 3D structures.

The "glyph field" visualizations (as seen in the composite above) show how the kernels effectively "flow" along orientations. This demonstrates their significant power for fiber enhancement in dMRI, where accurately following the direction of water diffusion is critical for mapping brain connectivity. The authors also mathematically proved how many convolutions are required to remove the initial "singularity" (the infinite spike at the starting point), ensuring numerical stability.

Future Directions

While these exact solutions represent a major milestone, they involve infinite series of complex functions, which can be computationally demanding for real-time applications. The authors suggest that while their new Gaussian approximation is an excellent "local" fix, further research is needed to bridge the gap between the speed of approximations and the perfect accuracy of exact solutions.

Future directions include connecting these kernels to sub-Riemannian geodesics—the "shortest paths" in these orientation-constrained spaces—and applying these tools to more complex "Brownian bridge" problems in 3D.

Citation: Portegies, J.M., & Duits, R. (2017). New Exact and Numerical Solutions of the (Convection-)Diffusion Kernels on SE(3). arXiv:1604.03843v6 [math.AP].

Summary updated

Abstract

We consider hypo-elliptic diffusion and convection-diffusion on $\mathbb{R}^3 \rtimes S^2$, the quotient of the Lie group of rigid body motions SE(3) in which group elements are equivalent if they are equal up to a rotation around the reference axis. We show that we can derive expressions for the convolution kernels in terms of eigenfunctions of the PDE, by extending the approach for the SE(2) case. This goes via application of the Fourier transform of the PDE in the spatial variables, yielding a second order differential operator. We show that the eigenfunctions of this operator can be expressed as (generalized) spheroidal wave functions. The same exact formulas are derived via the Fourier transform on SE(3). We solve both the evolution itself, as well as the time-integrated process that corresponds to the resolvent operator. Furthermore, we have extended a standard numerical procedure from SE(2) to SE(3) for the computation of the solution kernels that is directly related to the exact solutions. Finally, we provide a novel analytic approximation of the kernels that we briefly compare to the exact kernels.
BibTeX
@inproceedings{portegies2017new,
  author = {Jorg M. Portegies and Remco Duits},
  title = {{New Exact and Numerical Solutions of the (Convection-)Diffusion Kernels on SE(3)}},
  booktitle = {Differential},
  year = {2017},
  doi = {10.1016/j.difgeo.2017.06.004},
}
Figure