So far, it is still controversial about how to divide the liquid-like and gas-like states boundary and Widom delta, since the deviation of boundaries determined by different thermodynamic criteria increases significantly when away from the critical point. For this reason, a superior method that does not rely on thermodynamic criteria to define the boundary line between liquid-like and gas-like states needs to be proposed urgently and the recent extensive use of deep neural networks in molecular dynamics simulation provides a feasible choice. The present work attempts to determine a novel phase state boundary and Widom delta of supercritical CO₂ from extracting the microstructure features of subcritical vapor and liquid states in equilibrium by deep neural network coupled with molecular dynamics simulation. In addition, visualization of simulated systems containing distinct liquid-like molecule ratio π_LL of supercritical CO₂ is also presented as well as the radial distribution function at different states. The results show that the novel boundary is located in the middle of the Widom lines constructed by multiple thermodynamic criteria. The lower boundary T^- of the novel Widom delta agrees well with the theoretical boundary constructed by pseudo-boiling theory when it is close to the critical point but starts to deviate when it is far away from the critical point, while the upper boundary T⁺ of that is opposite. More importantly, the novel boundary and Widom delta constructed have no pressure upper limit compared to those constructed by thermodynamic criteria, only depending on π_LL, which means that the novel boundary and Widom delta can extend to the supercritical deep region. Further, the visualization and radial distribution function of simulated systems of supercritical CO₂ at different states provide the persistence of a liquid-like and a gas-like transition. The work conducted here can present novel microscopic insight into supercritical phase transition and provide another available alternative to define the Widom line.

迄今为止，对于如何划分类液体和类气体态边界以及Widom δ仍然存在争议，因为不同热力学准则所确定的边界的偏差在远离临界点时显着增大。因此，迫切需要提出一种不依赖热力学标准来定义类液体和类气体状态之间边界线的优越方法，而最近深度神经网络在分子动力学模拟中的广泛使用提供了一种可行的选择。目前的工作试图通过深度神经网络结合分子动力学模拟提取平衡状态下的亚临界汽态和液态的微观结构特征来确定超临界CO ₂的新相态边界和Widom delta 。此外，还展示了包含超临界CO _{2的不同类液体分子比率π LL的模拟系统的可视化}以及不同状态下的径向分布函数。结果表明，新边界位于由多种热力学准则构建的 Widom 线的中间。新维多姆三角洲的下边界T ^-在接近临界点时与伪沸腾理论构建的理论边界吻合较好，但在远离临界点时开始偏离，而上边界T ⁺与此相反。更重要的是，与热力学准则构建的边界和Widom三角洲相比，所构造的新边界和Widom三角洲没有压力上限，仅取决于π _LL，这意味着新边界和Widom三角洲可以延伸到超临界深部区域。_{此外，不同状态下超临界CO 2}模拟系统的可视化和径向分布函数提供了类液体和类气体转变的持续性。这里进行的工作可以提供对超临界相变的新颖的微观见解，并提供另一种定义 Widom 线的可用替代方案。