基于机器学习的合金催化剂表面吸附能预测

近年来，机器学习方法逐渐成为多相催化中的一种关键研究手段. 二元合金材料作为重要的催化剂之一，在双功能催化剂的筛选中受到了广泛的关注. 本文提出了一个将机器学习方法应用在预测催化性质上的整体框架，从而快速预测原子、分子在金属和二元合金表面的吸附能. 通过测试不同的机器学习方法来评估它们对于该问题的适用性，并将树集成的方法与压缩感知方法相结合，利用约6×10⁴个吸附能数据构建了预测模型. 相对于线性比例关系，该方法可以更准确地预测大量合金上的吸附能(预测的均方根误差降低一半)，并且更通用地预测各种吸附物的能量，为发现新的双金属催化剂铺平了道路.     

Fast and accurate modeling of molecular atomization energies with machine learning

We introduce a machine learning model to predict atomization energies of a diverse set of organic molecules, based on nuclear charges and atomic positions only. The problem of solving the molecular Schr?dinger equation is mapped onto a nonlinear statistical regression problem of reduced complexity. Regression models are trained on and compared to atomization energies computed with hybrid density-functional theory. Cross validation over more than seven thousand organic molecules yields a mean absolute error of ～10 kcal/mol. Applicability is demonstrated for the prediction of molecular atomization potential energy curves.     

A machine learning framework to predict the aggregation of polycyclic aromatic compounds

The physical aggregation of polycyclic aromatic compounds (PACs) is a key step in soot inception. In this work, we set out to elucidate which molecular properties of PACs influence the physical growth process and develop a machine learning framework to quantitatively relate these features to the propensity of PACs to physically dimerize. To this end, we identify a pool of compounds with a diverse range of properties and create a dataset of PAC monomers along with their calculated free energies of dimerization, obtained via molecular dynamics simulations enhanced by well-tempered Metadynamics. We then demonstrate that a machine learning model based on the least absolute shrinkage and selection operator (Lasso) is able to quantitatively learn how molecular features contribute to physical aggregation and predict the free energy of dimerization for new pairs of molecules. Results show that our model is able to accurately determine the stability of dimers obtained from both homo- and hetero-molecular dimerization cases. Our approach provides also a data driven method to determine the molecular features most important to predicting the dimer stability. Indeed, we identified size, shape, oxygenation, and presence of rotatable bonds as the most influential characteristics of PACs that contribute to physical dimerization. This work highlights the molecular complexity of the PAC monomers that must be accounted for in order to accurately represent physical aggregation. We anticipate that our approach is key to modeling soot inception as it allows for the efficient prediction of dimerization propensity from easily calculable molecular features.     

Machine learning approach for the prediction and optimization of thermal transport properties

Traditional simulation methods have made prominent progress in aiding experiments for understanding thermal transport properties of materials, and in predicting thermal conductivity of novel materials. However, huge challenges are also encountered when exploring complex material systems, such as formidable computational costs. As a rising computational method, machine learning has a lot to offer in this regard, not only in speeding up the searching and optimization process, but also in providing novel perspectives. In this work, we review the state-of-the-art studies on material’s thermal properties based on machine learning technique. First, the basic principles of machine learning method are introduced. We then review applications of machine learning technique in the prediction and optimization of material’s thermal properties, including thermal conductivity and interfacial thermal resistance. Finally, an outlook is provided for the future studies.     

The excitation energies,ionization potentials and oscillator strengths of neutral and ionized species of Uub (Z = 112) and the homologue elements Zn,Cd and Hg

Multi-configuration Dirac-Fock method (MCDF) is employed to calculate excitation energies, ionization potentials and oscillator strengths for all neutral and up to 5 times ionized species of element Uub, as well as the homologue elements Zn, Cd and Hg. On the basis of not too extended MCDF calculations, we studied some peculiar properties of element Uub resulting from its stronger relativistic and electron correlation effects. Using an extrapolative scheme, improved ionization potentials of Uub were obtained with an uncertainty of less than 0.5 eV. Furthermore, we calculated the low-lying resonance excitation energies, absorption oscillator strengths and the first ionization potential for Hg and Uub using large scale MCDF calculations, which improved the uncertainty of the excitation energies to less than 0.25 eV for element Hg. We hope that such calculations yield good results for element Uub.     

高效选取线性独立的原子特征构建精确机器学习势函数

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.     

Statistical properties of excited fissioning nuclei

The phenomenon of the disappearance of the shell effects on the thermodynamic properties of nuclei with increasing excitation energy has been examined quantitatively on the basis of numerical calculations based on realistic shell model single particle level schemes. It is shown that shell effects disappear at moderate excitation energies and above these excitation energies, the thermodynamic behaviour of the nucleus is identical to that of the equivalent liquid drop model nucleus. Implications of the above feature in the interpretation of some aspects of fission of excited nuclei such as mass-asymmetry and angular anisotropy are examined. The relationship of the phenomenon of washing out of shell effects at high excitation energies with the temperature smearing method of determining ground state shell correction energies is also outlined.     

基于CBM电子设备故障预测技术研究

为解决电子设备结构复杂,故障信息不足,故障预测困难,并且现有方法不能直接对电子设备进行状态预测等问题,本文提出了基于状态维修（CBM）的最小二乘支持向量机(LSSVM)和隐马尔科夫模型(HMM)组合故障预测方法。首先采取灵敏度分析法确定电路中要可能发生变化的元件,通过改变元件参数来设置电路的不同退化状态;其次建立组合故障预测模型;最后对该电路进行状态预测。结果表明,本文提出的方法能够直接预测电路的不同状态,进而实现直接预测电子设备的故障状态,预测精度可以达到93.3%。     

A Dynamic Anomaly Detection Approach Based on Permutation Entropy for Predicting Aging-Related Failures

Software aging is a phenomenon referring to the performance degradation of a long-running software system. This phenomenon is an accumulative process during execution, which will gradually lead the system from a normal state to a failure-prone state. It is a crucial challenge for system reliability to predict the Aging-Related Failures (ARFs) accurately. In this paper, permutation entropy (PE) is modified to Multidimensional Multi-scale Permutation Entropy (MMPE) as a novel aging indicator to detect performance anomalies, since MMPE is sensitive to dynamic state changes. An experiment is set on the distributed database system Voldemort, and MMPE is calculated based on the collected performance metrics during execution. Finally, based on MMPE, a failure prediction model using the machine learning method to reveal the anomalies is presented, which can predict failures with high accuracy.     

A Hybrid Method Based on Extreme Learning Machine and Wavelet Transform Denoising for Stock Prediction

The trend prediction of the stock is a main challenge. Accidental factors often lead to short-term sharp fluctuations in stock markets, deviating from the original normal trend. The short-term fluctuation of stock price has high noise, which is not conducive to the prediction of stock trends. Therefore, we used discrete wavelet transform (DWT)-based denoising to denoise stock data. Denoising the stock data assisted us to eliminate the influences of short-term random events on the continuous trend of the stock. The denoised data showed more stable trend characteristics and smoothness. Extreme learning machine (ELM) is one of the effective training algorithms for fully connected single-hidden-layer feedforward neural networks (SLFNs), which possesses the advantages of fast convergence, unique results, and it does not converge to a local minimum. Therefore, this paper proposed a combination of ELM- and DWT-based denoising to predict the trend of stocks. The proposed method was used to predict the trend of 400 stocks in China. The prediction results of the proposed method are a good proof of the efficacy of DWT-based denoising for stock trends, and showed an excellent performance compared to 12 machine learning algorithms (e.g., recurrent neural network (RNN) and long short-term memory (LSTM)).     

多任务神经网络对原子核低激发谱的研究

原子核低激发谱对深入理解原子核结构具有重要作用。采用多任务反向传播(Back Propagation,BP)的神经网络方法系统研究了原子核$ {2}_{1}^{+} $和$ {4}_{1}^{+} $的激发能量。除了质子数和中子数外,通过在网络输入层增加一个有关原子核集体性的物理量,BP神经网络在0.1 MeV到数MeV的能量范围内很好地拟合了原子核的低激发能。相比五维集体哈密顿量(Five-Dimensional Collective Hamiltonian,5DCH)方法,BP神经网络更好地再现了原子核低激发能的同位素趋势,以及由壳效应导致的幻数原子核低激发能的突然增大,并且将$ {2}_{1}^{+} $和$ {4}_{1}^{+} $激发能的预言精度分别提高了约80%和75%,该预言精度与单任务神经网络基本一致,但是改进了对轻核区与缺中子核区低激发谱的学习能力,这说明多任务神经网络可以实现多种激发能量的统一精确计算。     

电子与类锂离子碰撞激发

本文提出一种计算电子与离子碰撞激发截面的理论方法。同时,此方法也可作为一种检验和掌握电子碰撞激发数据的手段。在电子的低能区域,利用多通道量子数亏损理论,我们可从精确测得的能谱数据中推算出电子碰撞激发截面。在电子的高能区域,我们利用Bethe理论计算了电子碰撞激发截面。对中间能域,如果将截面的局部共振结构平均,则可以用内插法得到平均激发截面。比较可靠的电子与离子激发数据,对受控聚变的研究是有帮助的。本文以电子与类锂离子碰撞激发为实例说明所提出的理论方法。 关键词：     

有限深抛物势量子盘中极化子的激发态性质

量子点作为一种重要的低维纳米结构, 近年来在单光子光源和新型量子点单光子探测器的研究引起了人们的广泛关注, 对各种势阱中量子点性质的研究已取得了重要成果. 但是大多理论研究都局限于无限深势阱, 而有限深势阱更具有实际意义. 利用平面波展开、幺正变换和变分相结合的方法研究了有限深势阱中极化子激发态能量及激发能随势阱形状和量子盘大小的变化规律. 数值计算结果表明: 极化子的激发态能量、激发能随势垒高度或宽度的增大而增大, 原因是势垒愈高、愈宽, 电子穿透势垒的可能性愈小, 电子在阱内运动的可能性愈大, 进而导致极化子的激发态能量和激发能均随势垒高度和宽度的增大而增大; 极化子的激发态能量和激发能随量子盘半径的增大而减小, 表明量子盘具有显著的量子尺寸效应; 极化子的激发态能量随有效受限长度的增加而减小, 原因是有效受限长度愈大, 有效受限强度愈小, 电子受到的束缚愈弱、振动愈慢、势能愈小, 进而导致基态能量、激发态能量减小; 同时由于激发态能量较基态能量减小慢, 使得激发能随之增加. 研究结果对量子点的应用具有一定的理论指导意义.     

Production and state-selective detection of ultracold RbCs molecules

Using resonance-enhanced two-photon ionization, we detect ultracold, metastable RbCs molecules formed in their lowest triplet state a (3)Sigma(+) via photoassociation in a laser-cooled mixture of 85Rb and 133Cs atoms. We obtain extensive bound-bound excitation spectra of these molecules, which provide detailed information about their vibrational distribution, as well as spectroscopic data on several RbCs molecular states including a (3)Sigma(+), (2) (3)Sigma(+), and (1) (1)Pi. Analysis of this data allows us to predict strong transitions from observed levels to the absolute vibronic ground state of RbCs, potentially allowing the production of stable, ultracold polar molecules at rates in excess of 10(6) s(-1).     

基于高斯过程的混沌时间序列单步与多步预测

针对混沌时间序列单步和多步预测,提出基于复合协方差函数的高斯过程 (GP)模型方法.GP模型的确立由协方差函数决定,通过对训练数据集的学习,在证据最大化框架内,利用矩阵运算和优化算法自适应地确定协方差函数和均值函数中的超参数.GP模型与神经网络、模糊模型相比,其可调整参数很少.将不同复合协方差函数的GP模型应用在混沌时间序列单步及多步提前预测中,并与单一协方差函数的GP、支持向量机、最小二乘支持向量机、径向基函数神经网络等方法进行了比较.仿真结果表明,基于不同复合协方差函数的GP方法能精确地预测混沌时间序 关键词： 高斯过程 混沌时间序列 预测 模型比较     

Quantum chaos of bogoliubov waves for a bose-einstein condensate in stadium billiards

We investigate the possibility of quantum (or wave) chaos for the Bogoliubov excitations of a Bose-Einstein condensate in billiards. Because of the mean field interaction in the condensate, the Bogoliubov excitations are very different from the single particle excitations in a noninteracting system. Nevertheless, we predict that the statistical distribution of level spacings is unchanged by mapping the non-Hermitian Bogoliubov operator to a real symmetric matrix. We numerically test our prediction by using a phase shift method for calculating the excitation energies.     

Accelerating inverse crystal structure prediction by machine learning: A case study of carbon allotropes

Based on structure prediction method, the machine learning method is used instead of the density functional theory (DFT) method to predict the material properties, thereby accelerating the material search process. In this paper, we established a data set of carbon materials by high-throughput calculation with available carbon structures obtained from the Samara Carbon Allotrope Database. We then trained a machine learning (ML) model that specifically predicts the elastic modulus (bulk modulus, shear modulus, and the Young’s modulus) and confirmed that the accuracy is better than that of AFLOW–ML in predicting the elastic modulus of a carbon allotrope. We further combined our ML model with the CALYPSO code to search for new carbon structures with a high Young’s modulus. A new carbon allotrope not included in the Samara Carbon Allotrope Database, named Cmcm–C24, which exhibits a hardness greater than 80 GPa, was firstly revealed. The Cmcm–C24 phase was identified as a semiconductor with a direct bandgap. The structural stability, elastic modulus, and electronic properties of the new carbon allotrope were systematically studied, and the obtained results demonstrate the feasibility of ML methods accelerating the material search process.