Understanding melting behavior of aluminum clusters using machine learned deep neural network potential energy...

Area of Science:

• Computational materials science
• Physical chemistry
• Condensed matter physics

Background:

• Deep potentials (DP) offer near-density functional theory (DFT) accuracy for thermodynamic property calculations.
• Experimental data exists for smaller aluminum clusters, but finite-temperature ab initio simulations are lacking for larger sizes (N > 55).

Purpose of the Study:

• To perform finite-temperature ab initio simulations for aluminum clusters (N=48-342) to determine heat capacities and melting temperatures.
• To compare simulation results with existing experimental data and investigate discrepancies.
• To understand the influence of ground-state geometries and structural transitions on thermodynamic properties.

Main Methods:

• Employed the multiple histogram technique for heat capacity and melting temperature calculations.
• Conducted extensive molecular dynamics (MD) simulations across 24 temperatures for each cluster.
• Utilized structural and dynamical descriptors like mean squared displacements and the Lindemann index for analysis.

Main Results:

• Calculated heat capacities and melting temperatures for 32 aluminum clusters (N=48-342).
• Achieved good agreement with experimental melting temperatures for 19 clusters.
• Identified solid-solid structural transitions in Al116 and Al52, explaining observed heat capacity features.
• Found that ground-state geometries significantly influence heat capacity behavior.
• Investigated specific cluster anomalies (Al56, Al69, Al48) and attributed discrepancies to geometric changes or simulation limitations.
• Reported novel geometries and thermodynamics for 11 larger clusters (N=147-342).

Conclusions:

• Deep potential accuracy is reliable for MD simulations up to 750 K.
• Geometric factors, particularly for Al55, critically impact melting temperatures.
• Simulation results largely validate experimental findings, while also highlighting areas for further investigation and potential limitations of current models.