Learning Potential Energy Curves with Neural Networks and Lippmann–Schwinger Validation

We develop artificial neural-network (ANN) models that predict the potential energy as a function of interatomic separation for rare-gas dimers (Ar–Ar, Kr–Kr, Ne–Ne, Xe–Xe) and for the CO₂ dimer, using energy–distance data sampled from analytical reference curves. Hyperparameters—including layers, neurons, activation functions, and regularization—are optimized via multiple search strategies and evaluated across varied train/validation/test splits. We benchmark the ANN-predicted potentials against analytical reference curves and assess physical accuracy by computing dimer binding energies via the Lippmann–Schwinger integral equation. Across all systems, the ANN closely reproduces the reference curves and yields binding energies in good agreement with benchmarks.