Geometric features and a neural network classifier for detecting melting-like transitions in clusters.

Area of Science:

• Materials Science
• Computational Chemistry
• Statistical Mechanics

Background:

• Melting-like transitions in clusters are typically identified by peaks in heat capacity (C(T)) at the transition temperature (Tc).
• Calculating C(T) requires computationally expensive simulations with millions of steps, limiting efficiency.

Purpose of the Study:

• To introduce four easily calculable functions for detecting and characterizing melting-like transitions in clusters.
• To provide faster and more accurate estimates of melting points (Tc) compared to traditional methods.

Main Methods:

• Calculating the width of the potential energy distribution (WU).
• Analyzing statistics of interatomic distances (rij), including dissimilarity to the global minimum, interval counts around pair distribution peaks, and distribution non-uniformity.
• Utilizing an artificial neural network classifier to determine the solid fraction (FS(T)) and identify the solid-liquid coexistence region [Tf, Tm].

Main Results:

• The four proposed functions provide estimates for the melting region in good agreement with Tc across different bonding types (van der Waals, covalent, metallic).
• Inflection points in the third function and FS(T) serve as highly sensitive indicators of phase transitions.
• Estimates of Tc using these new methods converge 100 to 1000 times faster than those obtained via C(T).

Conclusions:

• The four novel functions, WU and interatomic distance statistics, offer an efficient alternative for detecting and characterizing cluster melting.
• Inflection points in specific functions and solid fraction calculations provide rapid convergence for phase transition temperature estimation.
• These methods significantly reduce simulation time, enhancing the feasibility of studying melting phenomena in materials.