Machine Learning Enabled Prediction of High Stiffness 2D Materials

Persistent
exploration of high stiffness two-dimensional (2D) materials
is necessary for advancements in applications such as nanocomposites,
flexible electronics, and resonant sensors, all of which demand elevated
resistance to deformation. However, data-centric material models developed
for this purpose remain in their early stages, often due to incomplete
stiffness estimation or limited transferability to unseen 2D materials.
In this context, we examined stiffness trends among different classes
of 2D materials and identified the elastic constants pivotal for estimating
the 2D material stiffness irrespective of their crystal symmetry.
Subsequently, we developed Gaussian Process Regression machine learning
models with the capability of relative stiffness comparison, which
are used to predict high stiffness candidates across a broad spectrum
of unseen 2D materials during model training. The probability of finding
high stiffness 2D materials increased significantly, from a mere 1%
in the training data set to a notable 47% in the set of machine learning-predicted
2D materials. We also discussed potential stiffening mechanisms, competing
stiffness characteristics, and complementary properties of these predicted
high-stiffness 2D materials that are crucial for enhancing the effectiveness
of the aforementioned applications.