120 Cu64Zr36 MG samples for GNN

This dataset is adapted for the dedicated Graph Neural Network (GNN) project (Link: https://github.com/WangChi-AiMat/GNN-MGDeform/)Sample Details:The primary dataset consists of 120 Cu64Zr36 metallic glass samples.Each sample contains exactly 6,480 atoms.The samples were generated at a cooling rate of 10^10 K/s.Simulation Process:All simulations were performed using LAMMPS, with Periodic Boundary Conditions (PBC) applied in all three spatial directions.For each MG sample, twelve fundamental athermal quasi-static (AQS) loading modes are applied. These include uniaxial tension and compression along the x, y, and z axes, as well as simple shear in the xy±, yz±, and xz± directions.During each deformation step, a small affine strain of 10^-4 is imposed. This is immediately followed by structural relaxation using conjugate gradient energy minimization. Because any complex loading scenario can be viewed as a linear combination of these twelve elementary modes, they provide a highly comprehensive measure of local plastic resistance.Target Prediction Variable:To measure the atomic propensity for plasticity, we evaluate the non-affine displacement squared (D²) at applied strains of 10%.Raw D² values naturally show a lognormal distribution (they are highly skewed). To make neural network training more stable and efficient, we use its natural logarithm, ln(D²).Finally, these ln(D²) values are averaged across all twelve fundamental loading modes and further normalized using the empirical cumulative distribution function, denoted as ECDF(ln(D²)).