A neuro-evolution machine learning potential for accurately describing defect-dependent mechanical behaviors in Mo–Nb alloys

In this work, a neuro-evolution machine learning potential integrated with the Ziegler–Biersack–Littmark repulsive potential is developed for accurately describing defect-dependent mechanical behaviors of molybdenum–niobium (Mo–Nb) alloys using only 721 training structures. Molecular dynamics simulations are then performed to investigate the individual and combined effects of Nb alloying and pre-existing voids on the uniaxial tensile deformation of Mo–Nb alloys. The developed potential accurately reproduces the structural, mechanical, and defect properties of Mo, Nb, and their alloys, in agreement with density functional theory calculations and experimental data, thus enabling reliable simulations of defect evolution and mechanical response. The molecular dynamics results indicate that increasing Nb content reduces peak strength but enhances the uniform deformation capability prior to strain softening by delaying shear-band nucleation, promoting multi-shear-band interactions, and suppressing localized deformation. Pre-existing voids degrade the tensile performance of the Mo matrix; larger voids cause earlier shear band initiation, faster propagation, and more severe localization, reducing both strength and plasticity. Adding Nb mitigates the detrimental effects of voids. In Mo–Nb alloys containing voids, a higher Nb content delays void-driven shear band evolution, restricts dominant shear band propagation, and promotes more uniform deformation. This work clarifies the deformation mechanisms of Nb solutes and voids, provides a reliable machine learning potential, and offers atomic-scale insights for optimizing Mo-based alloys.