Vibrational dynamics of TODD-graphene nanosheets via molecular dynamics simulations and machine-learning surrogate models: implications for adaptive nanomechanical devices.

contextTODD-Graphene (TODD-G), a porous two-dimensional carbon allotrope composed of interconnected 3-8-10-12 carbon rings, has recently emerged as a promising candidate for nanoelectromechanical and adaptive nanodevice applications due to its metallic conductivity, mechanical compliance, and topologically engineered phonon structure. In this work, the vibrational dynamics of TODD-G nanosheets were systematically investigated to establish structure-property relationships governing resonant behavior under varying geometric, thermal, mechanical, and defect conditions. Fundamental frequencies were found to span 2.93-13.28 THz and follow the classical

methodsAtomic-scale simulations were performed using classical molecular dynamics within the large-scale atomic/molecular massively parallel simulator (LAMMPS) package. Interatomic interactions were described using the adaptive intermolecular reactive empirical bond order (AIREBO) potential with activated Lennard-Jones and torsional terms, while multilayer van der Waals interactions employed a 12-6 Lennard-Jones potential. Structural relaxation was conducted using conjugate-gradient energy minimization. Thermal equilibration employed the Nosé-Hoover thermostat within the canonical (NVT) ensemble, followed by microcanonical (NVE) simulations for vibrational analysis using a timestep of 0.5 fs. Vibrational properties were extracted using impulse-response analysis, velocity autocorrelation functions, harmonic forcing, fast Fourier transform (FFT) spectral analysis, and phonon density-of-states calculations. Simulations systematically examined the effects of nanosheet size, temperature (1-1000 K), tensile strain (0-10%), vacancy-defect concentration (0.5-5%), multilayer stacking (1-5 layers), and boundary conditions (free-free, clamped-free, simply supported, and clamped-clamped). Machine-learning surrogate models including linear regression, support-vector regression with radial-basis-function kernels, random forest regression, and gradient-boosting regression were trained using an 80/20 train-test split and five-fold cross-validation to predict resonance frequency and quality factor across the multidimensional parameter space.