Analyzing Local Information in Trainable Materials with Internal Prestress

We model amorphous materials via disordered elastic network models to understand the complex mechanics of non-crystalline systems and their use as platforms for training in function. A useful feature of disordered materials is that local stresses of each bond can be stored as information which the material can use to tune itself for desired mechanical properties. This behavior has only been studied in the case when some realistic features of materials, such as internal stresses, are neglected. We find that the generalization to include internal stresses obscures the relevant local stress information needed for self-tuning. As currently practiced, tuning requires knowing the forces along each bond or strut – that is how much each bond is stretched or compressed. When there is internal stress, however, orientational motion of the bonds also becomes important. Therefore, information about only bond-lengths becomes less useful and information about perpendicular motion becomes important as pre-stress is added to the system. We show how both modes of deformation contribute to the mechanics and training algorithms.