Bayesian optimization-based design of defect gamma-graphyne nanoribbons with high thermoelectric conversion efficiency

Abstract

In this paper, we perform a systematical investigation on searching for defect γ-graphyne nanoribbons (γ-GYNRs) with optimal thermoelectric performance by utilizing nonequilibrium Green’s function combined with Bayesian optimization. The calculations show that Bayesian optimization could quickly and accurately identify the optimal structure with best thermoelectric efficiency. Even in the worst round of optimization, optimal structure could be obtained by only calculating 719 structure candidates (merely 4.35% of all the 16512 candidates). The room temperature thermoelectric figure of merit of optimal defect γ-GYNR (length 11.846 nm, width 1.453 nm) could approach 2.315, which is 5 times of that of pristine γ-GYNR. The obvious advantage of optimal defect γ-GYNR is mainly attributed to the maximum balance of weakening of thermal power factor (side effect) and suppression of thermal conductance (positive effect). Through analyzing the pair correlation function, we also find that the correlations between perfect (defect) unit and perfect (defect) unit are both extremely low for defect γ-GYNRs with high thermoelectric conversion efficiency. The findings presented in this work indicate the effectiveness and superiority of Bayesian optimization in designing defect γ-GYNRs, and could provide new insights for exploring other low-dimensional materials with excellent thermoelectric properties.