Differentiable graph-structured models for inverse design of lattice materials

Materials possessing flexible physico-chemical properties that adapt on-demand to the hostile environmental conditions of deep space will become essential in defining the future of space exploration. A promising venue for inspiration towards the design of environment-specific materials is in the intricate micro-architectures and lattice geometry found throughout nature. However, the immense design space covered by such irregular topologies is challenging to probe analytically. For this reason, most synthetic lattice materials have to date been based on periodic architectures instead. Here, we propose a computational approach using a graph representation for both regular and irregular lattice materials. Our method uses differentiable message passing algorithms to calculate mechanical properties, and therefore allows using automatic differentiation to adjust both the geometric structure and attributes of individual lattice elements to design materials with desired properties. The introduced methodology is applicable to any system representable as a heterogeneous graph, including other types of materials.