近年来, 基于机器学习的大规模原子模拟技术的发展为许多化学分支的发展带来了巨大的希望。这些模拟具有高速度和高精度的特点。本文概述了基于机器学习势函数的原子模拟的3个关键方面的最新进展, 即机器学习模型和结构描述子的开发, 全局势能面训练集的生成, 以及基于主动学习的势函数自动训练。研究表明, 最近设计的指数型结构描述子和前馈神经网络模型非常适合生成高度复杂的全局势能面。通过神经网络势函数在材料和反应模拟中的2个最新应用来说明基于机器学习势函数的原子模拟如何有助于发现新材料和反应。

针对目前指关节纹识别方法鲁棒性差的问题,提出了一种基于非下采样的Shearlet变换(NSST)和Tetrolet能量特征的指关节纹识别方法。首先,采用直方图均衡化调整图像的灰度,以减少光照分布不均对识别系统产生的影响。其次,利用NSST及其逆变换得到去噪后的重构图像,并对其进行Tetrolet变换,建立低频图像的能量曲面。最后,将不同图像的能量曲面作差,得到能量差曲面,进一步计算曲面的方差,并以此为依据对不同指关节纹图像进行分类识别。在HKPU-FKP、IIT Delhi-FK、和HKPU-CFK图库及其噪声图库的实验结果表明,本方法的正确识别率可达98.0392%,最短识别时间为0.0497 s,最低等误率为2.5646%。相比其他方法,本方法可以明显提高指关节纹识别系统的性能,具有可行性和有效性。

Atomic scale manufacturing is a necessity of the future to develop atomic scale devices with high precision. A different perspective of the quantum realm, which includes the tunnelling effect, leakage current at the atomic-scale, Coulomb blockade and Kondo effect, is inevitable for the fabrication and hence, the mass production of these devices. For these atomic-scale device development, molecular level devices must be fabricated. Proper theoretical studies could be an aid towards the experimental realities. Electronic transport studies are the basis to realise and interpret the problems happening at this minute scale. Keeping these in mind, we present a periodic energy decomposition analysis (pEDA) of two potential candidates for moletronics: phthalocyanines and porphyrins, by placing them over gold substrate cleaved at the (111) plane to study the adsorption and interaction at the interface and then, to study their application as a channel between two electrodes, thereby, providing a link between pEDA and electronic transport studies. pEDA provides information regarding the bond strength and the contribution of electrostatic energy, Pauli’s energy, orbital energy and the orbital interactions. Combining this analysis with electronic transport studies can provide novel directions for atomic/close-toatomic- scale manufacturing (ACSM). Literature survey shows that this is the first work which establishes a link between pEDA and electronic transport studies and a detailed pEDA study on the above stated molecules. The results show that among the molecules studied, porphyrins are more adsorbable over gold substrate and conducting across a molecular junction than phthalocyanines, even though both molecules show a similarity in adsorption and conduction when a terminal thiol linker is attached. A further observation establishes the importance of attractive terms, which includes interaction, orbital and electrostatic energies, in correlating the pEDA study with the transport properties. By progressing this research, further developments could be possible in atomic-scale manufacturing in the future.

本文应用准三体模型及扩展的LEPS势能面(PES)、准经典轨线和CPOAM模型计算了Ba+RBr(R=CH3,C2H5,C3J7,C4H9,n-C5H11)→BaBr+R反应体系产物BaBr的转动取向,结果表明产物BaBr的转动取向随碰撞能的增加越趋强烈,随烷基的增大而减弱.

本文利用代数方法研究了非对称弯曲三原子分子S₂O分子处于C～1A′电子态的能谱及其稳定构型下的势能面,通过对30条光谱数据的拟和得到的RMS误差为2.40 cm-1.结果表明,利用此代数Hamiltonian很好的实现了能级再现,它预测了振动总量子数达到20的全部振动能级(在本文中我们只列举到v=9),同时我们计算了分子的解离能与力常数.通过与实验值比较证明了这种方法在计算这类分子的有效性.

利用量子力学耦合通道扭曲波近似法(CCDWA)和三种势能面计算了H+H2碰撞的反应几率,结果发现在相同的势能面下利用CCDWA方法计算的反应几率和公认较好的计算结果符合很好,不同势能面共线势垒高度的差别将引起反应几率的不同.

应用三体模型及扩展的LEPS势能面(PES),对初始条件为(Ecol=55kJ/mol,v=0,j=0)的O(1D)+N2O→NO+NO反应体系进行了准经典轨线(QCT)计算.根据计算结果对体系的势能面及反应机理进行详细的分析和讨论,较全面地研究了此反应体系的动力学特征.