Interactive Visualization of Large Scale Atomistic and Cosmological Particle Simulations

摘要

In this paper we present a method for interactive highquality visualization of large and time-varying 3D unstructured data from particle simulations. To help design next-generation materials for aerospace systems, understanding the physics of material structures at the atomic scale is achieved through molecular dynamics simulations run on tens to hundreds of millions of atoms. Similarly, to understand the formation of the cosmos, galaxy formation simulations are run using tens to hundreds of millions of particles. From the atomically small to the astronomically large, these simulations produce immense amounts of data. While these simulations have been shown to produce excellent results matching realworld observations, the daunting Big Data that they produce can be difficult to work with. Numerical analysis is useful for processing this large quantity of data, but visual analysis is also important in the search for identifying interesting results and formulating new hypotheses to explore. Though many level-of-detail techniques exist for structured and geometric data, allowing scalable interactive visualization, unstructured particle data is not suitable for these techniques. Therefore, we have developed a new technique for scalable visualization of massive particle datasets. The technique is different from previous level-of-detail techniques which create progressive approximations of the data itself. Instead, the technique creates a volumetric representation of the data first and then creates a hierarchy of progressive approximations of the volumetric representation. An adaptive volume rendering technique is also presented that progressively refines the volume hierarchy depending on the view of the data and the time available to load data. The technique provides a scalable framework, allowing various levels of detail to be loaded while maintaining interactivity. This paper describes the rendering techniques used to produce the 3D volume hierarchy as well as the adaptive volume renderer used for interactive visualization. Three case studies are discussed as examples of interactive visualization. Two are from a molecular dynamics simulation, one containing approximately 22 million particles per time-step over 100 time-steps and one containing a range of particles containing a total of 230 million particles over 9 timesteps. The third example is a single timestep containing 128 million particle from a cosmology simulation.