Design and visualization of Riemannian metrics

Tiago Novello, Vinícius da Silva, Luiz VelhoarXiv 2020 · arXiv

Introduction

Traditional computer graphics typically operate under the assumption of Euclidean geometry, where light travels along perfectly straight lines. However, simulating phenomena such as gravitational lensing, atmospheric mirages, or stylized spatial warps requires a shift to non-Euclidean frameworks. In these spaces, light follows geodesics—paths of shortest distance governed by the underlying Riemannian metric. This paper introduces a novel framework for real-time ray tracing in such spaces by constructing custom Riemannian metrics within $\mathbb{R}^3$.

Methodology: Real-Time Geodesic Ray Tracing

The core challenge in non-Euclidean rendering is the computational cost of calculating geodesics. The authors formulate ray tracing as solving the geodesic flow on a manifold. Instead of a single straight vector, rays are modeled as polygonal chains integrated step-by-step using numerical methods.

Geodesic flow on a Riemannian manifold $M$, where light paths $\gamma(t)$ are determined by the metric at point $p$.

To make this interactive, the paper proposes two tractable methods for metric design:

  1. Graphs of Functions: By treating 3D space as a pullback of a 4D graph $w = f(x,y,z)$, the authors derive explicit formulas for Christoffel symbols using only the gradient and Hessian of $f$.
  2. Diffeomorphisms: Using a smooth mapping $\Phi$, the authors present a novel mathematical proof (Theorem 5.1) that provides a direct way to compute Christoffel symbols from the Jacobian and Hessian of $\Phi$, bypassing complex tensor calculus.

These deformations are compositional, allowing for the creation of complex environments by summing functions (e.g., multiple Gaussian "bumps") or composing mappings using the chain rule.

Visualization and Results

The framework was evaluated using procedurally generated 3D grids to visualize the effects of various spatial deformations. By leveraging the simplified mathematical formulations and GPU-accelerated numerical integration via Euler's method, the authors achieved real-time performance on modern Nvidia RTX hardware.

Visual comparison of a standard Euclidean grid (left) against spaces warped by local Gaussian "bumps" (center) and global diffeomorphic twists (right).

The results demonstrate that ray tracing is uniquely suited for these geometries. In certain curved spaces, multiple light paths can connect the same two points—a condition where traditional rasterization fails due to the non-injective nature of the exponential map.

Discussion and Future Work

A significant contribution of this work is the simplification of the mathematical pipeline for GPU implementation. Theorem 5.1, in particular, enables technical artists to program complex deformations without deep expertise in differential geometry.

However, limitations remain. The current ray-scene intersection test is performed sequentially for each segment of the curved ray, which is inefficient. Future optimizations could involve parallelizing these tests to better utilize GPU hardware. Additionally, while the Euler method provides the speed necessary for real-time interaction, higher-order integration methods like Runge-Kutta could improve accuracy over long distances where integration drift might occur.

Summary updated

Abstract

Local and global illumination were recently defined in Riemannian manifolds to visualize classical Non-Euclidean spaces. This work focuses on Riemannian metric construction in $\mathbb{R}^3$ to explore special effects like warping, mirages, and deformations. We investigate the possibility of using graphs of functions and diffeomorphism to produce such effects. For these, their Riemannian metrics and geodesics derivations are provided, and ways of accumulating such metrics. We visualize, in "real-time", the resulting Riemannian manifolds using a ray tracing implemented on top of Nvidia RTX GPUs.
BibTeX
@misc{novello2020design,
  author = {Tiago Novello and Vinícius da Silva and Luiz Velho},
  title = {{Design and visualization of Riemannian metrics}},
  year = {2020},
  eprint = {2005.05386},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2005.05386},
}
Figure