Unifying mechanics and thermodynamics: The atomic-scale origin of stability in titanium alloys

The thermomechanical stability of high-performance materials such as titanium alloys is essential to their functional reliability, yet predicting this stability is challenging due to the complex interplay between local stress and thermodynamic driving forces at the atomic scale. Here, using large-scale molecular dynamics simulations, we establish a direct, quantitative correlation between local von Mises stress, configurational entropy, and Helmholtz free energy in representative α - and β -phase titanium alloys. We reveal that β -phases exhibit broader stress-entropy distributions, indicating enhanced atomic mobility and reduced thermodynamic stability. Crucially, we resolve a common ambiguity by demonstrating that α -phases, despite having comparable per-atom energies, possess lower free energy per unit volume, which confirms their superior thermodynamic stability at low to moderate temperatures. We further show that the geometric constraints of the HCP α -phase lattice produce a much stronger coupling between local stress and free energy than in the more deformable BCC β -phase. These insights are integrated into a machine learning model that accurately predicts atomic-scale thermodynamic states of the alloy across a wide temperature range. Our findings provide an interpretable, atomistically resolved framework for predicting stress-driven disorder and phase stability in complex engineering alloys. • Volumetric free-energy analysis confirms α -phase titanium alloys are thermodynamically more stable than β -phases at moderate temperatures. • The constrained HCP crystal structure of the α -phase results in a significantly stronger coupling between free energy and mechanical stress compared to the more deformable β -phase. • Bivariate maps linking atomic entropy and local stress provide a quantitative fingerprint of a material’s thermomechanical state. • An interpretable machine-learning surrogate, trained on statistically sufficient MD data, accurately predicts atomic-scale states from temperature, enabling rapid alloy screening. • The established link between structure and stability provides a transferable framework for accelerating data-driven alloy design.