Author(s): Abreu, C. S. ; Gomes, J. R.
Date: 2025
Persistent ID: https://hdl.handle.net/1822/94855
Origin: RepositóriUM - Universidade do Minho
Subject(s): Machine learning; Molecular dynamics; Friction and wear; Carbon nanotubes; Deep neural network learning potential
Accelerating the discovery of materials with desired properties has been a long-standing challenge in materials sciences. Nonetheless, the computational cost of accurate quantum-mechanical calculations proves prohibitive for simulations involving larger spatial and time scales. In recent years, increased efforts to overcome this bottleneck have been undertaken using machine learning (ML), where only a reduced set of reference calculations is required to accurately predict potential-energy surfaces. Atomistic Molecular Dynamics (MD) simulations will, therefore, greatly benefit from such pathway to achieve more accurate predictions of properties in relation to empirically parameterized potentials, namely in the study of tribological systems. Nanofabricated motors have found increasing importance in the NanoElectroMechanical Systems (NEMS) realm. However, current microfabrication techniques cannot produce surfaces smooth enough at the coupling between rotor and bearing, leading to relevant energy dissipation and reduction in rotating speeds. Multiwalled carbon nanotubes (MWCNT) can overcome such shortcomings owing to strong and intrinsically atomically smooth surfaces, enabling easy inter-shell sliding. In this work, the friction and wear performance of MWCNT nanobearings is assessed using a Deep Neural Network Potential (DNNP) model trained for energies and forces of MD trajectories. The deep ML force field was trained against MD22 benchmark dataset using SchNetPack 2.0 framework interfaced with LAMMPS classical MD code. Inner and outer nanotubes with lengths of 6 nm and 5 nm, respectively, were set in relative motion with rotational frequencies varying in the leeway 25-250 GHz. To capture all relevant dynamics, an integration timestep of 1 fs was used so as to be an order of magnitude smaller than the timescale of the fastest phonons.
