Atomic structure of liquid refractory Nb₅Si₃intermetallic compound alloy based upon deep neural network potential

The knowledge of atomic structure for liquids, especially for liquid alloys with complex crystal structure and high liquidus temperatures, remains poorly understood. In this work, we have extended the development of deep neural network (DNN) potential for liquid Nb5Si3. The DNN potential captures the structural features of liquid alloys compared with ab initio results. The chemical short-range order parameter suggests that there exists strong affinity between Nb and Si atoms. The dynamic property was investigated, and the diffusion coefficient obeys the Arrhenius relationship. The atomic structure has been subsequently explored for normal and undercooled liquid Nb5Si3. Large amounts of fivefold symmetry Honeycutt-Andersen pairs have been identified in liquid Nb5Si3. However, due to the violent thermal motion in a high-temperature Nb5Si3 melt, icosahedral symmetry and distorted icosahedrons (ICOs) account for little proportion according to Voronoi polyhedron (VP) analysis. The effect of thermal motion on VPs has been discussed. Except from the well documented «0,2,8,2»and «0,1,10,2»distorted ICOs, six more quasi-ICOs («0,1,9,3», «0,2,8,1», «0,2,8,4», «0,2,8,5», «0,1,10,3», and «0,1,10,4») have been proven to deform from ICOs at high temperatures. The local environment motif obtained by the atomic cluster alignment method demonstrates the existence of dominant distorted ICOs. At last, the atomic structure during melting process is discussed by VP analysis. It is found that «0,2,8,1», «0,2,8,2», «0,2,8,5», and «0,1,10,4»prefer to form at the beginning of the melting but rapidly reduce when it is fully melted.