高效选取线性独立的原子特征构建精确机器学习势函数（英文）

机器学习势由于具有与第一性原理计算相当的准确性,且低得多的计算成本,在原子模拟中极具前景.然而原子机器学习势的可靠性、速度和可迁移性在很大程度上取决于原子构型的表示.适当地选取用作机器学习程序输入的描述符是一个成功的机器学习表示的关键.本文发展了一种简单有效的方法,可以基于训练数据固有的相关性,从大量待选的描述符中自动选取一组最佳的线性独立原子特征.通过对几个具有较少冗余线性独立嵌入密度描述符的基准分子构建嵌入原子神经网络势的应用,证明了这种新方法的有效性和准确性.该算法可以大大简化原子特征的初始选取,并极大地提高原子机器学习势的性能.     

用修正嵌入原子法（MEAM）研究金属铝的稳定性

用两种不同的嵌入原子法（即ＥＡＭ和ＭＥＡＭ）计算了Ａｌ的结合能、层错能，并考察了它的稳定性，发现用修正嵌入原子法计算出的结果与已有的实验和理论结果基本一致，作者认为对立方晶系材料，原子的电子密度可用自由原子外层电荷密度的线性迭加表示，而不用修正嵌入原子法中的解析拟合过程表示。     

表面Cu原子间相互作用对Cu(001)表面跳跃扩散行为的影响

采用嵌入原子方法的原子间相互作用势，利用准静态分子动力学模拟研究了Cu原子在Cu(001)表面吸附所导致的基体晶格畸变以及对其附近的另一个吸附原子自扩散行为的影响.研究结果表明，吸附原子的存在可以导致多达10层的Cu基体晶格产生畸变.两个吸附原子所产生的晶格畸变应力场之间的相互作用，可以导致吸附原子运动活性的增加.通过比较同一路径上往返跳跃扩散势垒的差异发现，在原子间相互作用势的有效距离之外，两个吸附原子的扩散行为可以认为是存在晶格畸变应力场相互作用的两个独立吸附原子的扩散；在原子间相互作用势的有效距离之 关键词： 表面吸附原子 晶格畸变 表面二聚体 扩散     

紧束缚势和修正的嵌入原子势计算液态Sn双体相关函数差异

对500个Sn原子分别用两种模型(紧束缚势和修正的嵌入原子势)计算了400℃～1700℃温度范围内纯Sn的双体相关函数g(r).将计算结果与实验数据进行了对比分析,发现两种计算结果都能基本上反映液态Sn的结构及其随温度的变化情况:原子最近邻距离与实验结果相近;随着温度降低,双体相关函数第一峰变得尖锐,第二峰变得明朗.修正的嵌入原子势模型得到的双体相关函数的第一峰右侧有个突起的肩膀,这在实验结果中也被发现,而紧束缚势模型得到的双体分布函数肩膀不明显.     

Ni,Al及其合金的一种嵌入原子势

嵌入原子势在金属材料的结构及其物性的计算机模拟方面仍然有着重要的作用.针对面心立方结构的镍、铝及其合金,我们拟合了一种简单形式的嵌入原子势.势参数的拟合使用了相应材料的晶格常数,结合能、空位形成能以及三个弹性常数C_(11),C_(12),C_(44).除了一些高频部分的合理偏差之外,使用拟合的势参数得到的声子谱与实验结果符合良好.此外,得到的状态方程也与第一原理的理论结果很好地符合,说明了此嵌入原子势的可靠性.     

数字信号在原子芯片中的应用

通过分析和计算不同谐波分量与原子相互作用产生不同的势场,发现可以将其叠加在一起形成原子囚禁势,提出了数字信号在原子芯片中的应用方案. 关键词： 原子芯片 数字信号 绝热势     

核反应堆压力容器模拟钢中富Cu原子团簇的析出与嵌入原子势计算

利用原子探针层析技术(APT)和热处理时效方法,研究了合金元素Ni对核反应堆压力容器模拟钢中富Cu原子团簇析出的影响.实验结果表明,添加合金元素Ni(0.84wt%)的样品中析出富Cu原子团簇的数量密度高于不添加Ni的样品,富Cu原子团簇内以及团簇和基体界面处都有Ni元素的富集现象,这说明合金元素Ni会促使富Cu原子团簇的析出.从多体势的角度出发,利用嵌入原子势理论,基于纯金属元素Fe,Cu,Ni的多体势参数,建立了Fe-Cu二元和Fe-Cu-Ni三元体系的嵌入原子多体势.计算结果表明,当模拟合金中存在1at%Ni时有利于富Cu原子团簇的析出,这与实验结果相符.     

惰性气体原子间相互作用势比较研究

原子间相互作用势是预测惰性气体输运性质的必要输入条件. 文章对描述惰性气体原子间相互作用的Lennard-Jones势、指数排斥势、Hartree-Fock-Dispersion-B (HFD-B)势和唯象势的形式和特点进行了分析. 基于Chapman-Enskog方法, 计算得到了惰性气体在300–5000 K温度区间内基于四种原子相互作用势的黏性和热导率, 并与文献报道的实验和理论计算结果进行了比较. 研究结果表明, 基于Hartree-Fock排斥理论与色散理论发展起来的HFD-B势能够合理反映惰性气体原子相互作用的趋势与特征, 因而可以较好地预测惰性气体的宏观输运性质.     

紧束缚势和修正的嵌入原子势计算液态Sn双体相关函数差异

对500个Sn原子分别用两种模型（紧束缚势和修正的嵌入原子势）计算了400℃~1700℃温度范围内纯Sn的双体相关函数g（r）。将计算结果与实验数据进行了对比分析,发现两种计算结果都能基本上反映液态Sn的结构及其随温度的变化情况：原子最近邻距离与实验结果相近;随着温度降低,双体相关函数第一峰变得尖锐,第二峰变得明朗。修正的嵌入原子势模型得到的双体相关函数的第一峰右侧有个突起的肩膀,这在实验结果中也被发现,而紧束缚势模型得到的双体分布函数肩膀不明显。     

嵌入原子法计算金属钚中点缺陷的能量

钚因放射性衰变而出现老化效应.钚中点缺陷的性质和行为是理解钚老化效应的一个基础和前提.运用分子动力学模拟技术,计算了金属钚中点缺陷和点缺陷团簇的形成能和结合能.其中钚-钚、钚-氦和氦-氦相互作用势分别采用嵌入原子多体势、Morse对势和Lennard-Jones对势.计算结果表明,单个自间隙原子易以〈100〉哑铃状形态存在;间隙氦原子在理想晶格的八面体间隙位置相对较为稳定;氦原子与空位的结合能较大,在钚的自辐照过程中两者易于结合并形成氦-空位团簇;氦-空位团簇的形成能随氦原子数的增加而增大,当氦与空位的数     

Thermodynamic and physical properties of FeAl and Fe3Al: an atomistic study by EAM simulation

With this work we present a newly developed potential for the Fe–Al system, which is based on the analytical embedded atom method (EAM) with long range atomic interactions. The potential yields for the two most relevant phases B2-FeAl and D0₃-Fe₃Al lattice constants, elastic constants, as well as bulk and point defect formation enthalpies, which are in good agreement with experimental and other theoretical data. In addition, the phonon dispersions for B2-FeAl and D0₃-Fe₃Al show a good agreement with available experiments. The calculated lattice constants and formation enthalpy for disordered Fe–Al alloys are in good agreement with experimental data or other theoretical calculations. This indicates that the present EAM potentials of Fe–Al system is suitable for atomistic simulations of structural and kinetic properties for the Fe–Al system.     

机器学习法的干旱区典型农作物分类

当前,基于机器学习方法开展农作物分类研究,对于确保干旱区粮食安全和生态安全有着极为重要的现实意义。基于机器学习方法,采用时间序列Sentinel 2A遥感数据提取农作物分类信息,通过引入地块基元和红边特征,探讨了不同分类特征组合对机器学习分类精度的影响。结果表明：随机森林分类器可以有效集成光谱和植被指数等多维向量的优势,将其应用于干旱区典型农作物分类上的精度均在89%以上,分类组总体精度最高可达94.02%。地块基元点集支持下的分类特征提取方法能够提高机器学习效率和农作物分类精度,使光谱组及指数组的分类精度分别提高3.13%和4.07%,并能有效解决“椒盐”效应及耕地边缘廓线模糊等问题。红边光谱和红边指数的引入分别使随机森林分类器总体精度提高2.39 %和1.63%,并使春、冬小麦的识别能力显著提高,表明红边特征能够帮助分类器更敏感地捕捉不同作物特有的生长特性及物候差异。该研究结果可为机器学习方法及Sentinel 2A卫星在干旱区农业遥感的应用提供参考。     

从单分子到凝聚相水体系的理论描述：势能面和分子动力学最新进展

本文回顾了近十年来水体系的势能面与分子动力学理论研究的最新进展,包括水分子参与的气相反应,固体表面上的吸附与解离动力学,以及从团簇到凝聚相水的结构、振动光谱与统计力学模拟. 近年来再次发展起来的机器学习技术,例如结合置换不变多项式的神经网络,或结合基本不变量的神经网络,已被成熟应用于气相与固体表面体系的高精度势能面构造中. 对于团簇甚至凝聚相水体系,原子中心神经网络方法或基于核的高斯过程方法应用更为广泛. 此外,在多体展开框架下,在气相体系中发展起来的的方法也组成了高维度体系势函数构造的高精度方案. 当前凝聚相水体系面临的主要问题是高精度从头算数据集的积累,兼顾计算精度与效率的双杂化密度泛函是一种可能的解决方案. 在动力学理论方面,无论是化学反应截面计算还是振动光谱模拟,往往需要合理描述水分子中氢原子的量子效应,才能得到较为可靠的理论计算结果. 量子波包动力学方法已经在气相反应机理研究方面有深入的应用,也在包含数个水分子的团簇振动分析中有初步应用. 基于路径积分的分子动力学方法正在较大水团簇以及凝聚相水的结构与谱学模拟方面发挥重要作用.     

Dislocation modeling in bcc lithium: A comparison between continuum and atomistic predictions in the modified embedded atoms method

In this study, the modified embedded-atom method (MEAM) was applied to compare the predictions of dislocation core properties obtained by molecular statics with the continuum predictions obtained in the framework of the simplified 1D-Peierls–Nabarro model. To this end, a set of four fictive Li potentials in the MEAM framework was proposed with the condition that all four potentials reproduce the same elastic constants, the same transition energies between bcc and fcc crystal structures, and between bcc and hcp crystal structures, while the unstable stacking fault energy on the plane {110} in the direction <111> was varied around the value predicted by first-principles. Within these potentials, direct atomistic calculations were performed to evaluate dislocation core properties such as dislocation half width and Peierls stress and the results were compared with continuum predictions. We found that the trends predicted by the Peierls–Nabarro model, i.e. (i) a decrease of the dislocation half width with increasing unstable stacking fault energy, and (ii) an increase of the Peierls stress with increasing the magnitude of the unstable stacking fault energy, were recovered using atomic calculations in the MEAM framework. Moreover, the magnitude of the dislocation half width and the Peierls stress calculated in the MEAM framework are in good agreement with the Peierls–Nabarro predictions when the dislocation half width is determined using a generic strategy. Specifically, the dislocation half width is defined as the distance for which the disregistery is included between b/4 and 3b/4. It was, therefore, demonstrated herein that the set of fictive potentials could be parameterized in the MEAM framework to validate or to disprove the continuum theory using atomistic methods.     

Machine Learning to Instruct Single Crystal Growth by Flux Method

Growth of high-quality single crystals is of great significance for research of condensed matter physics. The exploration of suitable growing conditions for single crystals is expensive and time-consuming, especially for ternary compounds because of the lack of ternary phase diagram. Here we use machine learning(ML) trained on our experimental data to predict and instruct the growth. Four kinds of ML methods, including support vector machine(SVM), decision tree, random forest and gradient boosting decision tree, are adopted. The SVM method is relatively stable and works well, with an accuracy of 81% in predicting experimental results. By comparison,the accuracy of laboratory reaches 36%. The decision tree model is also used to reveal which features will take critical roles in growing processes.     

MEAM study of carbon atom interaction with Ni nano particle

Carbon, Ni, and C-Ni alloy modified embedded atom method (MEAM) potentials were developed to study the initial process of carbon nanotube growth on Ni catalyst particles. The MEAM potentials were used to study the atomistic interaction between a carbon atom and a fcc Ni nano particle, both on the particle surfaces and inside the Ni nano particles. The result shows that surface carbon atom is more stable than those in the bulk and sub-surface interstitial positions. Carbon atoms are expected to diffuse from the bulk to the surface, and the single walled and double-walled carbon nanotubes would be more favorable to form on Ni nano particle catalyst. The carbon and Ni nano particle interaction calculation shows that the corner and the edge of the particle are the energetically more favorable sites for the carbon adatom. The carbon nanotube may grow from the corner and edge of the particle.     

Statistical model of an atom in electron scattering calculations

The parameters describing scattering of electrons by atoms, which usually involve multiple integration of the atomic potentials, may be severely affected by the accuracy of these potentials. In the present work an iterative procedure is proposed providing a sequence of solutions of the Thomas-Fermi equation with increasing accuracy. Such a sequence makes it possible to establish the sensitivity of a given parameter to the accuracy of the atomic potential, and consequently to determine the accurate value of this parameter. Based on the present solutions, the differential scattering cross-sections for the Thomas-Fermi atom are calculated, and are found to deviate from the literature data.     

How machine learning conquers the unitary limit

Machine learning has become a premier tool in physics and other fields of science.It has been shown that the quantum mechanical scattering problem cannot only be solved with such techniques,but it was argued that the underlying neural network develops the Bom series for shallow potentials.However,classical machine learning algorithms fail in the unitary limit of an infinite scattering length.The unitary limit plays an important role in our understanding of bound strongly interacting fermionic systems and can be realized in cold atom experiments.Here,we develop a formalism that explains the unitary limit in terms of what we define as unitary limit surfaces.This not only allows to investigate the unitary limit geometrically in potential space,but also provides a numerically simple approach towards unnaturally large scattering lengths with standard multilayer perceptrons.Its scope is therefore not limited to applications in nuclear and atomic physics,but includes all systems that exhibit an unnaturally large scale.