Deep Learning Molecular Dynamics Study Of Optical And Thermal Transport Properties Of Liquid Water

Water is widely found in nature and industry,and its optical and thermodynamic properties play an important role in marine remote sensing,energy dynamics,planetary exploration,and biochemistry.However,due to its mobility and selective absorption properties,the measurement bands of the optical properties of water are limited,and the temperature dependence is difficult to observe.Also,due to the complexity of the intermolecular microstructure of water,the optical and thermodynamic properties of water are still not fully known and understood by using theoretical computational methods to study them.In order to provide basic data support for theoretical models and industrial applications of water,we evaluated the optical and thermodynamic properties of liquid water using molecular dynamics simulations on the basis of empirical force fields,density general function theory,and machine learning potential.The effects of temperature,size effect,and accuracy of potential function on infrared spectra and thermal conductivity are explored.The leading research results of this paper are as follows:(1)The optical properties of water were investigated based on the molecular dynamics approach.We both look into how temperature affects the infrared spectrum of water,which was computed using first principles molecular dynamics simulations.The findings indicate that the peak intensity of the OH stretching band changes most significantly.With the increase of temperature,its infrared absorption intensity gradually decreases and its spreading width becomes smaller,which is closer to the experimental value.The static dielectric parameter of water(87.41)was calculated based on the SPC/E water model,which is basically in agreement with the experimental value(78.54).(2)The thermodynamic properties of water were studied based on the classical molecular dynamics method.Using the equilibrium molecular dynamics method,the heat conductivity of three common water models(TIP3P,SPC/E,and TIP4P/2005)was calculated and compared.The computed results showed that the heat conductivity values calculated for the SPC/E water model were closer to the experimental values.It is obvious that when the temperature rose,the heat conductivity did as well.Toinvestigate the effects of water model type and system size,we used the non-equilibrium molecular dynamics method,and the results demonstrated that the thermal conductivity of water tended to remain constant with increasing system size when the system length reached 5.588 nm,and the effect of size effect was negligible,indicating that the results reached convergence.When the two methods’performance in calculating thermal conductivity was compared,it was discovered that the nonequilibrium molecular dynamics method’s result was closer to the experimental value.(3)This is a passage discussing the use of deep neural network methods to construct a machine learning potential with quantum accuracy for liquid water and to explore its optical properties.We collected data in the temperature range of 300-360 K using first-principles molecular dynamics and trained a high-precision deep learning potential using deep neural networks.The deep learning potential has excellent predictions for system energy and atomic forces and can well reproduce the radial distribution function calculated by the first-principles molecular dynamics,indicating that it can study the structural changes and dynamic properties of liquid water with quantum mechanical accuracy.We used the deep learning potential model to calculate the infrared spectrum of liquid water and found that there are three major peaks,located at 723.2043 cm-1,1624.4282 cm-1,and 3326.7380 cm-1,respectively.Compared with experimental data,the peak position offsets were 108.9842 cm-1,-15.7318 cm-1,-64.472 cm-1,respectively,with small errors and good consistency with experimental data.The frequency-dependent dielectric function was calculated according to the Debye relaxation model,and it was found that the real part of the frequency-dependent dielectric function has a small difference from experimental data in the low-frequency region,and at the same time the imaginary part can reproduce some experimental data well.