JAX, M.D. A framework for differentiable physics*

We introduce JAX MD, a software package for performing differentiable physics simulations with a focus on molecular dynamics. JAX MD includes a number of physics simulation environments, as well as interaction potentials and neural networks that can be integrated into these environments without writing any additional code. Since the simulations themselves are differentiable functions, entire trajectories can be differentiated to perform meta-optimization. These features are built on primitive operations, such as spatial partitioning, that allow simulations to scale to hundreds-of-thousands of particles on a single GPU. These primitives are flexible enough that they can be used to scale up workloads outside of molecular dynamics. We present several examples that highlight the features of JAX MD including: integration of graph neural networks into traditional simulations, meta-optimization through minimization of particle packings, and a multi-agent flocking simulation. JAX MD is available at https://www.github.com/google/jax-md.

Sound damping in soft particle packings: the interplay between configurational disorder and inelasticity

We numerically investigate sound damping in disordered two-dimensional soft particle packings. Our findings suggest that sound damping in soft particle packings is determined by the interplay between elastic heterogeneities and inelasticity.

Strain dependent vorticity in sheared granular media

Displacement fields in sheared particle packings often display vortex-like structures that reveal essential features about the mechanical state of the collection of particles. There are several metrics to quantify these flow field features, yet extracting such quantitative metrics from flow field or particle tracking data involves making numerous choices on the time and length scales over which to average. Here we employ a much used experimental data set on sheared disk packings to explore how such arbitrary data mining choices affect the obtained results. We focus on calculating the strain dependent vorticity, as this metric is a differential method hence potentially sensitive to the way it is computed. We find that the total surface area with an absolute vorticity above a certain threshold approaches a plateau value as shear progresses. This plateau value exhibits a non-monotonic dependence on packing fraction. We also show which range of choices yields results that can support an analysis method independent, physical interpretation of the flow field data.

Transparent experiments: releasing data from mechanical tests on three dimensional hydrogel sphere packings

We describe here experiments on the mechanics of hydrogel particle packings from the Behringer lab, performed between 2012 and 2015. These experiments quantify the evolution of all contact forces inside soft particle packings exposed to compression, shear, and the intrusion of a large intruder. The experimental set-ups and processes are presented and the data are concomitantly published in a repository (Barés et al. in Dryad, Dataset 10.5061/dryad.6djh9w0x8, 2019).