含能晶体密度预测的研究进展

密度是决定含能材料爆轰性能的重要参数。为评估现有CHON类含能材料密度的计算方法,对等电子密度面法、分子表面静电势法、基团加和法、晶体堆积法、定量构效关系法、经验公式法等进行分析和归类。结果表明,基于分子体积预测方法的精度取决于分子间和分子内相互作用对密度影响描述的准确度。其中,准确描述氢键和van der Waals作用充满了挑战性。基于晶体体积计算密度的核心在于晶体结构的准确预测,结构搜索要面对巨大的状态空间和高度复杂的能量曲面的困难,预测效率是亟待解决的问题。体积加和法和经验公式法存在无法区分同分异构体和晶型的缺点,且对新发现的具有特殊结构的分子由于缺乏实验数据难以获得准确的经验参数,计算结果偏差较大。引入人工神经网络、遗传算法以及支持向量机等机器学习算法后,定量构效关系法在含能化合物性能与结构关系研究中取得很大成就,模型精度进一步提高将为基于材料基因组模式的含能材料设计研发奠定基础,这也是今后密度预测方法发展的主要方向。

Review of Crystal Density Prediction Methods for Energetic Materials

Crystal density is an important parameter for predicting the detonation performance of energetic materials (EMs). Many studies have shown that the theoretical calculation methods are able to figure out accurate densities of CHNO contained EMs. In this work, we overview and categorize some reliable crystal density calculation methods, including isosurface of electron density method, group addition method, molecular surface electrostatic potentials method, crystal packing method and quantitative structure-property relationship method. Among these methods, the effectiveness of molecular volume-based methods depends on its capability to estimate inter- and intramolecular interactions. It is challenging to accurately describe the hydrogen bonding and van der Waals interactions. Due to the huge structure group spaces and highly complex potential energy surface, the crystal packing methods based on empirical forcefields are computationally expensive and lacking accuracy usually. The group addition approach cannot distinguish conformers and polymorphs, and may be unreliable for novel or special energetic materials, which are absent from accurate empirical parameters. The disadvantage of quantitative structure-property relationship method is that it is difficult to give the physical meaning of the equation. The bottleneck of insufficient experimental data and poor model accuracy needs to be solved. Nevertheless, numerous artificial intelligence methods, such as artificial neural networks, genetic algorithm, multiple linear regression, machine learning, have made great achievements in the relationship between properties and structure, facilitating the development of energetic materials based on the materials genome concept and serving as a main tendency in future.