Structure And Thermal Transport Properties Of Disordered Carbon And Silicon By Machine Learning Molecular Dynamics Simulations

Disordered carbon and silicon materials have wide applications and promising development opportunities in energy storage,wear-resistant coatings,solar cells,and microelectronic devices.It is elemental to understand the structural and thermal transport properties of both disordered materials toward to their practical applications.Due to the factors such as long-range disorder and diversity of structures,it is difficult to achieve experimentally quantitative characterizations or accurate measurements for these disordered materials.Molecular dynamics(MD)simulation,as one theoretic computational method on the atomic scale,can give insights into the atomistic structure and heat transport of these disordered materials.The interatomic potential dictates the reliability of MD results.Unfortunately,because of the poor quality,traditional empirical potentials fail to describe these disordered systems.In this thesis,we have developed Gaussian approximation potential(GAP)and neuroevolution potentials(NEP)for disordered carbon and silicon over the accurate and efficient machine learning(ML)techniques of Gaussian process regression and neural evolution.Through ML potential driven melt-quench largescale MD simulations,we have comprehensively studied density-dependent structural and thermal transport properties of nanoporous carbon(NP-C)and amorphous carbon(a-C),as well as the structural properties with different quenching rates and temperature-dependent thermal conductivities of amorphous silicon(a-Si).Key contents as follows:(1)We have retrained GAP-ML potential with van der Waals interaction and prepared 131072-atom NP-C samples in the mass density range of 0.5-1.7 g/cm3(sample size,L=18-12 nm)by melt-graphitization-quench MD profile.By means of pairwise correlation function,angular distribution function,coordination statistics,ring counts,X-ray diffraction and pore characterization,we have thoroughly analyzed structural properties and nano pore distributions of NP-C with different densities.These results show the density-independent characteristics of both short-and medium-range order,while the pores give a Gaussian-like distribution,whose peak positions and widths vary from the density,and their pore diameter can reach up to 4 nm.These calculations are essentially consistent with some of available experimental results.(2)NEP-ML potential model with comparable accuracy to GAP is trained for disordered carbon.Through the rapid melt-quench MD profile,we have prepared 125000-atom a-C samples with the density range of 1.5-3.5 g/cm3(L=12-9 nm).The density-dependent short-and medium-range structure properties of a-C are comprehensively characterized by means of pairwise correlation function,angular distribution function,coordination statistics and static structure factor.And some of these results are in good agreement with experiments.(3)We have retrained NEP-ML potential and obtained 64000-atom(L=11 nm)a-Si samples with different quenching rates by low melt-quench MD method.Using similar characterization techniques as mentioned above,we conclude that the slowest quenching rate of α=1011 K/s makes the most realistic a-Si sample.(4)By means of homogeneous non-equilibrium MD and thermal spectral decomposition,we have computed density-dependent thermal transport properties of a-C(NP-C)in the range of 1.5-3.5 g/cm3(0.3-1.5 g/cm3)at room temperature in detail.The calculated density-dependent thermal conductivity of a-C is consistent with the experimental measurements via quantum statistical correction applied on spectral thermal conductivity.(5)Combining heterogeneous and homogeneous non-equilibrium MD with related thermal spectral decomposition methods,we have calculated the temperature(from 10 to 1000 K)-and thickness-dependent thermal conductivities of a-Si prepared by slowest quenching rate.After quantum statistical correction based on the spectral thermal conductivity,we accurately achieved the prediction of thermal conductivity of a-Si over a large temperature range from 10 K to room temperature.By means of ML-driven MD calculation,the structure and thermal transport properties of disordered carbon and silicon have been comprehensively studied.We hope that these results can provide basic theoretical guidance for the understanding of structure of disordered covalent materials and their applications in future.