Post-Gaussian variational method for quantum anharmonic oscillator

合集下载

风云四号红外高光谱GIIRS中波通道亮温偏差订正

风云四号红外高光谱GIIRS中波通道亮温偏差订正

文章编号:1672-8785(2021)05-0039-06风云四号红外高光谱GIIRS中波温王根12陈娇1戴娟3王悦1$.安徽省气象台大气科学与卫星遥感安徽省重点实验室,安徽合肥230031;2.中亚大气科学研究中心,新疆乌鲁木齐830002;3.气,230031)摘要:变分同化风云四号干涉式大气垂直探测仪(Geostationary InterferometricInfrared Sounder,GIIRS)中波通道亮温偏差高,需进行GI-IRS资料偏差&在Harns B A等人提出的“离线”法的,了基于随机(Random Forest,RF)的GIIRS偏差法&在行过程中,基于风云四号多通道扫描成像辐射计(Advanced Geosynchronous Radiation Imager,AGRI)云产品对GIIRS资料进行了测&,经过偏差的GI-IRS亮温偏差高的。

与“离线”法,RF法的效好。

关键词:高光谱GIIRS;偏差订正;“离线”法;随机森林;云检测中图分类号:P407文献标志码:A DOI:10.3969/j.issn.1672-8785.2021.05.007BiasC0rrecti0n0fBrightne s Temperaturesin Medium WaveChannelof FY-4A Infrared Hyperspectral GIIRSWANG Gen GH,CHEN Jiao',DAI Juan3,WANG Yue1(9.Anhui Key Lab of Atmospheric Science and Satellite Remote Sensing,Anhui MeteorologicalObservatrry,Hefei230031,China;2.Center of Central Asia Atmospheric ScienceResearch,Urumqi830002,China;3.Anhui Climate Center,Hefei230031,China) Abstract:The brightne s temperature bias of the medium wave channel of the variational a s imilation geosta­tionary interferometrc infrared sounder(GIIRS)of F5-4is required to meet the Gaussian clistnbution,so thebias correction of GIIRS data is necessary.Based on Harns B A and Kelly Gs"off-line"method,a method forG I RSbiasco r ectionbasedontherandomforestisdevelopedinthispaper.Inthespecificimplementationproce s theclouddetectionofG I RSdataisca r iedoutbasedontheadvancedgeosynchronousradiationima-ger(AGRI)cloudproductsofFY-4.Theexperimentalresultsshowthatthebrightne s temperaturebiasofG I RS satisfies the assumption of Gau s ian distribution after the bias co r pared with"o f-line"method randomforestmethodhasabe t erco r ectione f ect.收稿日期:2021-01-07基金项目:国家自然科学基金项目(41805080);中亚大气科学研究基金项目(CAAS202003);安徽省气象台自立项目(AHMO202007;AHMO202004)作者简介:王根(1983-),男,江苏泰州人,高级工程师,博士,主要从事卫星资料同化、正则化反问题与人工智能应用等方面的研究&E-mail:203wanggen@Key words:hyperspectral GIIRS;bias correction;"offline"method;random forest;cloud detection数值天气预报是一个初/边值问题&星载高光谱红外探测器通道主要覆盖CO2和HQ 光谱区域&CO?和HQ吸收带提供的温度和湿度值预报的模式变量。

gaussian的介绍

gaussian的介绍

F、第三周期以后的原子的基组
第三周期以上的原子的基组很难处理。 由于存在非常大的核,原子核附近的电子通 过有效核电势方法(ECP)进行了近似,这一处 理同时也包含了相对论效应。这其中, LANL2DZ 是最有名的基组。
常用基组总结:
应用原子 基组 [H-Xe] STO-3G [H-Xe] 3-21G [H-Cl] 6-31G(d)(6-31G*) 6-31G(d, p) (6-31G**)[H-Cl] [H-Cl] 6-31+G(d) [H-Cl] 6-31+G(d,p) [H-Br] 6-311+G(d,p) [H-Br] 6-311+G(2d,p) [H-Br] 6-311+G(2df,2p) [H-Br] 6-311++G(3df,2pd) 描述与说明
C、超SCF方法:考虑一些电子相关的理论有很多
种。一般的,这些方法被称为超SCF方法,因为都是在 Hartree-Fock方法上增加了电子相关的因素。 Moller-Plesset 微扰由二级到五级的关键词是MP2, MP3,MP4,MP5。提供优化方法的有 MP2 和 MP3, MP4(不包括MP4SDQ),频率分析提供 MP2。在密度泛函 方法得到广泛应用之前,MP2 方法是考虑电子相关的最便 宜的方法,它可以成功应用于很多领域,一般都能得到很精 确的结果,是理论化学中非常有力的工具。在实践上,只有 MP4得到比较广泛的应用。MP3对于 MP2处理不好的体系 一般也没有好的结果。 二次 CI,一般包括三重和四重QCISD,QCISD(T), QCISD(TQ),QCISD提供优化。
E、高角动量基组
现在使用的更大的基组,是在分裂基组基础上增加多个 角动量。比如 6-31G(2d)就是在6-31G基础上增加两个 d 轨道 的函数,而 6-311++G(3df,3pd)则增加了更多的极化函数, 包括三个分裂的价键基组,在重原子和氢原子上加的弥散函 数,在重原子上加的三个 d 函数和一个 f 函数,在氢原子上 加的三个 p 函数和一个 d 函数。这样的基组在电子相关方法 对于描述电子之间的作用有很重要意义。这些基组一般不用 于HF计算。 一些大的基组根据重原子的周期数而增加不同的极化函 数。如 6-311+(3df,2df,p)基组在第二周期以及以上都采用 三个 d 函数和一个 f 函数的极化,而对于第一周期采用两个 d 函数和一个 f 函数的极化。注意一般从头算所说的周期是没 有氢原子所在的周期的。即碳处于第一周期。

基于混合高斯变分自编码网络的异常检测算法

基于混合高斯变分自编码网络的异常检测算法

基于混合高斯变分自编码网络的异常检测算法陈华华,陈哲,郭春生,应娜,叶学义,章坚武(杭州电子科技大学通信工程学院,浙江杭州 310018)摘 要:异常数据是指偏离大量正常数据点的数据,往往会对各类系统产生负面影响,存在较大风险。

异常检测作为一种有效的防护手段,能够检测数据中的异常,为各类系统的正常运转提供重要支撑,具有重要的现实意义。

提出了一种基于混合高斯变分自编码网络的异常检测算法,该算法首先使用混合高斯先验构建变分自编码器,对输入数据进行特征提取,然后以混合高斯变分自编码器构建深度支持向量网络,压缩特征空间,并寻找最小超球体分离正常数据和异常数据,通过计算数据特征到超球体中心的欧氏距离衡量数据的异常分数,并以此进行异常检测。

最后在基准数据集MNIST和Fashion-MNIST上评估了该算法,平均AUC分别达到了0.954和0.937。

实验结果表明,所提出的算法取得了较好的异常检测效果。

关键词:异常检测;变分自编码器;混合高斯分布;超球体中图分类号:TP393文献标识码:Adoi: 10.11959/j.issn.1000−0801.2021044Anomaly detection algorithm based onGaussian mixture variational auto encoder networkCHEN Huahua, CHEN Zhe, GUO Chunsheng, YING Na, YE Xueyi, ZHANG Jianwu School of Communication Engineering, Hangzhou Dianzi University, Hangzhou 310018, China Abstract: Anomalous data, which deviates from a large number of normal data, has a negative impact and contains a risk on various systems. Anomaly detection can detect anomalies in the data and provide important support for the normal operation of various systems, which has important practical significance. An anomaly detection algorithm based on Gaussian mixture variational auto encoder network was proposed, in which a variational autoencoder was built to extract the features of the input data based on Gaussian mixture distribution, and using this variational au-toencoder to construct a deep support vector network to compress the feature space and find the minimum hyper sphere to separate the normal data and the abnormal data. Anomalies can be detected by the score from the Euclidean distance from the feature of data to the center of the hypersphere. The proposed algorithm was evaluated on the benchmark datasets MNIST and Fashion-MNIST, and the corresponding average AUC are 0.954 and 0.937 respec-tively. The experimental results show that the proposed algorithm achieves preferable effects.Key words: anomaly detection, variational autoencoder, Gaussian mixture distribution, hypersphere收稿日期:2020−08−05;修回日期:2021−03−06·55·电信科学 2021年第4期1 引言异常检测是指在数据中检测出不符合期望行为的数据。

英文简历写法样本

英文简历写法样本

Curriculum VitaeProfessor Shaher MomaniDepartment of Mathematics,Mutah UniversityP.O.Box7,Al-Karak,Jordan.Mobile:+962(0)777500326E-mail:shahermm@,smmom@.joWeb Site:.jo/userhomepages/shmomani/index.htmWeb Site:.jo/jrgam/index.htmlPersonal Data•Full Name:Shaher Mohammad Ahmad Momani•Date of Birth:May10,1962•Place of Birth:Ajloun-Jordan•Nationality:Jordanian•Sex:Male•Marital Status:Married,two daughters and two sons•Profession:Professor of Mathematics at Mutah UniversityAcademic Qualification•Ph.D.in Mathematics,Applied Mathematics(Non-Newtonian Fluid Mechanics) University:University:of Wales,United Kingdom,1991.Title of Thesis:Some Problems in Non-Newtonian Fluid Mechanics.Advisor:Professor Ken Walters.•B.Sc.in MathematicsUniversity:Yarmouk University,Jordan,1984.Experience•Editor-in-Chief:Arab Journal Of Mathematics And Mathematical Sciences(AJMMS).Web Site:/ajmms.htm•Leader of Jordan Research Group in Applied Mathematics(JRGAM).Web Site:.jo/jrgam/index.html•Professor:Mutah University,September20067to present.1•Professor:Qatar University,September2006to September2007.•Associate Professor:Mutah University,September2004to September2006.•Associate Professor:Jordan University,Summer course2004.•Associate Professor:United Arab Emirates University,September2001to September2004.•Associate Professor:Yarmouk University,September2000to September2001.•Associate Professor:Mutah University,November1998to September2000.•Assistant Professor:Mutah University,September1991to November1998.•Head of Department of Mathematics,Mutah University,September1994to September1995.•Visiting Professor:Department of Mathematics,University of Wales,June-August1996,UK. Research InterestsMy general research interests are in the areas of applied mathematics,Non-Newtonian Fluid Mechanics, differential equations of applied mathematics,fractional calculus and fractional differential equation. More specifically,my research interests can be summarized as follows:1.Numerical solution of ordinary and partial differential equations of fractional order.2.Theory of fractional differential equations and integral equations.3.Newtonian and Non-Newtonianfluid mechanics.4.Stability of fractional linear systems.5.Fractional chaotic systems.6.Variational inequalities and obstacle problems.7.Mathematical Modelling.8.Mathematical Physics.9.Solitary Waves.10.Oil Shales.Editor1.Arab Journal of Mathematics and Mathematical Sciences,(Editor-in-Chief).Website:/ajmms.htm2.International Journal of Nonlinear Dynamical Systems and Chaos(IJNDSC).Website:/ijndsc1.htm3.World Journal of Modelling and Simulation(WJMS).Website:/4.Journal of Fractional Calculus and Its Applications.25.International Journal of Computational Mathematics and Numerical Simulation(IJCMNS).Website:/ijcmns/othereditors2.htmWebsite:/6.Journal of Nonlinear and Fractional Phenomena in Science and Engineering.Website:/index.htm7.Applied Mathematics&Information SciencesWebsite:/8.International Review of Pure and Applied Physics.9.Mutah Journal for Scientific Research.Website:.jo/dar/arabic/abstracs/edibo05s.htmReviewer•Journal of Computational and applied Mathematics.•Journal of Mathematical Analysis and Applications.•Physics Letters A.•Physica Scripta.•Electronic Journal of Differential Equations.•Communications in Nonlinear Science and Numerical Simulation.•Chaos,Solitons&Fractals.•Journal of Applied Analysis.•Arabian Journal for Science and Engineering.•Computers and Mathematics with Applications.•International Journal of Computer Mathematics.•Journal of Applied Mathematics.•International Journal of Mathematics and Mathematical Sciences.•Numerical Methods for Partial Differential Equations.•Referee for several international and local journals.Membership1.Member of the International Who’s Who,since2006.2.Member of the Jordanian Mathematics Society,since1991.3.Member of the Exactive Committee of the Jordanian Mathematics Society,1994-1996.4.Member of the British Society of Rheology,since1989.5.Member of Newton Institute of Non-Newtonian Fluid Mechanics.University of Wales.36.Member of the American Mathematical Society,since1996.7.Leader of Jordan Research Group in Applied Mathematics(JRGAM),Jordan,2005−Present.8.Member of the fractional calculus and its applications cuminuty.http:http://www.tuke.sk/podlubny/fc.html9.Member of the Research Group in Mathematical Inequalities and Applications(RGMIA). Academic Honors and Grants1.The Islamic Educational,Scientific and Cultural Organization Science Prize”ISESCO SciencePrize”(2008).2.Third World Academy of Sciences(TWAS)prize,in Italy,(2000).3.A Scholarship from Mutah University to get a PhD(1987-1991).4.A Grant from Qatar University to do a Research Project on Chaotic Dynamics and Phase Syn-chronization of Fractional Order Dynamical Systems(2007-2008).5.Classified as One of the Top Ten Scientists in the World in Fractional differential Equationsaccording to Thomoson ISI.6.The Elsevier Prize for Jordan Scientists(2009).Some of my Co-workers1.Professor Ishak Hashim,School of Mathematical Sciences,University Kebangsaan Malaysia.2.Professor Shijun Liao,Shanghai Jiao Tong University,China.3.Professor Hossain Erjaee,University of Qatar,Qatar.4.Professor Nabil Shawagfeh,University of Jordan,Jordan.5.Professor S.B.Hadid,Ajman University for Science and Technology,UAE.6.Professor Ahamed Alawaneh,University of Jordan,Jordan.7.Professor Muhammad Aslam Noor,COMSTATS Institute of Information Tech.,Pakistan.8.Professor Dogan Kaya,Firat University,Turkey.9.Dr Kamel Al-Khaled,Jordan University of Science and Tech.,Jordan.10.Dr Reyad Al-Khazali,Etisalat College,UAE.11.Dr Zaid Odibat,Al-Balq’a University,Jordan.12.Dr Hossein Jafari,Mazandaran University,Babolsar,Iran.13.Dr Vedat Suat Erturk,Department of Mathematics,Ondokuz Mayis University,Turkey.14.Dr Rabha Ibrahim,Sana’a,Yemen.4Committee Service•Mutah University1.Member of the Organizing Committee for the2nd Jordanian Mathematics Conference,1994.2.Member of the Organizing Committee for the3rd Jordanian Mathematics Conference,1996.3.Member of the Organizing Committee for A Workshop in Teaching Calculus Using Mathe-matica Software,Mutah Univ.,1996.4.Member of a Committee to Study the Academic Status at Mutah Univ.5.Member of the Housing Committee1995-1996.6.Member of the Library Committee1995-1996.7.Member of the Supervising Committee of Mutah Univ.School1995-1997.8.Member of Several Internal Committees in the Faculty and Department.9.Head of the Committee for Establishing the Master Program at Mutah University,2000.10.Member of the Seminar Committee2005-present.11.Member of the Graduate Studies Committee2005-present.12.Member of the Appointing Committee of Mutah Univ.School2005.13.Member of Several Internal Committees in the Faculty and Department.14.Member of Several Committees at Mutah University during the academic years2004-2006.15.Coordinator and Member of Several Committees at Mutah University during the academicyears2007-present.5•Yarmouk University1.Member of the Scientific Research Committee.2.Member of Several Internal Committees in the Faculty and Department.•United Arab Emirates University1.Coordinator of the Practical Training and Field Trips Committee.2.Coordinator of the Counseling and Advising Committee.3.Member of the Cultural and Social Committee.4.Member of the Information and Annual Report Committee.5.Member of the Curricular Committee in the Faculty.6.Member of Several Internal Committees in the Faculty and Department.•Qatar University1.Coordinator and Member of Several Internal Committees in the Faculty and Department.•Outside Universities1.Member of the Committee for Maths.Department of Irbid Private Univ.2.Member of the Committee for Maths.Department of Zarqa Private Univ.3.Member of the Committee for Maths.Department of Ziatounah Private Univ.4.Coordinator of the Committee for Maths.Department of Jadara Private Univ.5.Referee for Research Papers Publish in Various Journals.6.Member of a Defense exam Committee for Several Master Theses in Jordan Universities.7.Chaired a Session of the Third Jordanian Mathematics Conference,1996.8.Chaired a Session of the Fifth Annual U.A.E.University Conference,2004.9.Chaired a Session of the Recent Advances in Mathematics Conference,India,2004.10.Member of the Exactive Committee of Qualification Exam at Jordan Universities.11.Chaired a Session of The Third Conferences On Research And Education In Mathematics,Malaysia2007.12.Member of the Organizing Committee of The2nd International Symposium on NonlinearDynamics,Shanghai,China,2007.13.Chaired a Mini-Symposium in The2nd International Symposium on Nonlinear Dynamics,Shanghai,China,2007.14.Member of the International Program Committee of the Fractional Differentiation and itsApplications,Ankrah,Turkey,2008.Computer Skills1.Mathematical Software:Mathematica,Fortran,Maple,Matlab.2.Operating Systems:DOS,UNIX,Windows2000.3.Typesetting Software:Tex,LaTeX,Scientific Workplace,MicroSoft Word.4.MCDL:Mutah Computer Drive License.6Published and Accepted PapersThis is a complete list of the papers of mine which either have appeared or have been accepted for publication.Owing to copyright restrictions,the papers are not available in electronic form.If you are interested in any of these papers,send me an e-mail and I will send you a reprint.1.G.Georgiou,Shaher Momani,M.J.Crochet,and K.Walters,Newtonian and non-Newtonianflow in a channel obstructed by an antisymmetric array of cylinders,Journal of Non-Newtonian Fluid Mechanics,Vol.40,(1991)231-260.2.S.Hadid,B.Maseadeh and Shaher Momani,On the existence of maximal and minimalsolutions of differential equations of non-integer order,Journal of Fractional Calculus,Vol.9, (1996)41-44.3.S.B.Hadid, A.A.Ta’ani and S.M.Momani,Some existence theorems on differentialequations of generalized order through afixed-point theorem,Journal of Fractional Calculus,Vol.9,(1996)45-49.4.B.S.Maseadeh,A.A.Ta’ani and S.M.Momani,On w-compact spaces,J.of Institute ofMathematics and Computer Sciences,(1996).5.B.Maseadeh,S.Momani and S.Hadid,Solutions of differential equations of non-integerorder in L2and C spaces,Mutah Journal for Research and Studies,Vol.12(1),(1997)169-181.6.Shaher Momani,Theflow of non-Newtonianfluids through corrugated pipes,Mu’tah Journalfor Research and Studies,Vol.12(4),(1997)91-112.7.S.Momani and S.Hadid,An algorithm for numerical solutions of fractional order differentialequations,Journal of Fractional Calculus,Vol.15,(1998)61-66.8.S.M.Momani and K.Walters,Theflow of non-Newtonianfluids through curved pipes,Al-Dirasat Journal,Vol.26(1),(1999)74-87.9.S.M.Momani,On existence of solutions of a system of ordinary differential equations of frac-tional order,Far East Journal of Mathematical Sciences(FJMS),Vol.1(2),(1999)265-270. 10.S.M.Momani,Variation of solutions of differential equations of non-integer order with respectto initial condition and parameters,Far East Journal of Mathematical Sciences(FJMS),Vol.1(3),(1999)423-428.11.S.M.Momani,Stress distribution and pressure gradient of non-Newtonianfluids through con-verging ducts,Mu’tah Lil-Buhooth Wa Al-Dirasat Journal,Vol.15(1),(2000)9-26.12.S.M.Momani,Local and global uniqueness theorems on differential equations of non-integerorder via Gronwall’s and Bihari’s inequalities,Revista Technica Journal,Vol.23(1),(2000)66-69.13.Shaher Momani,Numerical solution of differential equations of non-integer order by the gener-alized difference method,Al-Zarqa Private University Journal,Vol.2(1),(2000)1-7.14.S.M.Momani,On the existence ofε-approximate solutions of differential equations of non-integer order,PanAmerican Mathematical Journal,Vol.10(3),(2000)61-69.15.Shaher M.Momani,Local and global existence theorems on fractional integro-differential equa-tions,Journal of Fractional Calculus,Vol.18,(2000)81-86.16.Shaher M.Momani and Reyad El-Khazali,On the existence of extremal solutions of frac-tional integro-differential equations,Journal of Fractional Calculus,Vol.18,(2000)87-92.717.S.M.Momani,Theflow of non-Newtonianfluids through rotating pipes,Al-Manara Journal,Vol.7(1)(2001),9-25.18.Shaher Momani,Some existence theorems on fractional integro-differential equations,AbhathAl-Yarmouk Journal,Vol.10(2B),(2001)435-444.19.S.M.Momani and S.B.Hadid,Asymptotic behaviour of the maximal and minimal solutionsof differential equations of non-integer order,Far East Journal of Mathematical Sciences(FJMS), Vol.6(1),(2002)31-39.20.S.M.Momani and S.B.Hadid,Dependence of solutions of differential equations of non-integer order on initial conditions and parameters,Al-Manara Journal,Vol.9(2),(2003)69-76.21.Reyad El-Khazali,Shaher Momani,Stability analysis of composite fractional systems,In-ternational Journal of Applied Mathematics,Vol.12(1),(2003)73-85.22.S.M.Momani and S.B.Hadid,On the inequalities of integro-differential fractional equations,International Journal of Applied Mathematics,Vol.12(1),(2003)29-37.23.S.M.Momani and S.B.Hadid,Some comparison results for integro-fractional differentialinequalities,Journal of Fractional Calculus,Vol.24,(2003)37-44.24.S.M.Momani,S.B.Hadid and Z.M.Alawaneh,Some analytical properties of solutionsof differential equations of the noninteger order,International Journal of Mathematics and Math-ematical Sciences,Vol.2004(13),(2004)697-701.25.Shaher Momani,Analytical solutions of strongly non-linear oscillators by the decompositionmethod,International Journal of Modern Physics C(IJMPC),Vol.15(7),(2004)967-979.26.Shaher Momani and Samir Hadid,Lyapunov stability solutions of fractional integro-differential equations,International Journal of Mathematics and Mathematical Sciences,Vol.2004(47),(2004)2503-2507.27.Shaher Momani and Kamel Al-Khaled,Numerical solutions for systems of fractional dif-ferential equations by the decomposition method,Applied Mathematics and Computation,Vol.162(3),(2005)1351-1365.28.S.M.Momani and S. B.Hadid,On the continuous dependence of solutions of integro-fractional differential equations with respect to initial conditions,Nonlinear Functional Analysis and Applications,Vol.10(3),(2005)379-386.29.Kamel Al-Khaled and Shaher Momani,An approximate solution for a fractional diffusion-wave equation using the decomposition method,Applied Mathematics and Computation,Vol.165(2),(2005)473-483.30.Shaher Momani,Analytical approximate solution for fractional heat-like and wave-like equa-tions with variable coefficients using the decomposition method,Applied Mathematics and Com-putation,Vol.165(2),(2005)459-472.31.Shaher Momani,Khaled Moadi and Muhammad Aslam Noor,Modified decompositionmethod for solving a system of third-order obstacle problems,International Journal of Pure and Applied Mathematics,Vol.21(1),(2005)97-107.32.Shaher Momani,An explicit and numerical solutions of the fractional KdV equation,Mathe-matics and Computers in Simulation,Vol.70(2),(2005)110-118.833.Shaher Momani,A numerical scheme for the solution of Sivashinsky equation,Applied Mathe-matics and Computation,Vol.168(2),(2005)1273-1280.34.Shaher Momani,Analytic and approximate solutions of the space-and time-fractional telegraphequations,Applied Mathematics and Computation,Vol.170(2),(2005)1126-1134.35.Kamel Al-Khaled,Shaher Momani and Ahmed Alawneh,Approximate wave solutions fora generalized Benjamin-Bona-Mahoy-Burgers equation,Applied Mathematics and Computation,Vol.171(1),(2005)281-292.36.Muhammad Aslam Noor,S.K.Mishra and Shaher Momani,Properties of approximatepreinvex functions,Nonlinear Analysis Forum Journal,Vol.10(2),(2005)1-9.37.Shaher Momani and Salah Abuasad,Application of He’s variational iteration method toHelmholtz equation,Chaos,Solitons&Fractals,Vol.27(5),(2006)1119-1123.38.Zaid Odibat and Shaher Momani,Application of variational iteration method to nonlineardifferential equations of fractional order,International Journal of Nonlinear Science and Numer-ical Simulation,Vol.7(1),(2006)27-34.39.Shaher Momani,Non-perturbative analytical solutions of the space-and time-fractional Burgersequations,Chaos,Solitons&Fractals,Vol.28(4),(2006)930-937.40.Shaher Momani,Khaled Moadi and Muhammad Aslam Noor,Decomposition methodfor solving fourth order obstacle problems,Applied Mathematics and Computation,Vol.175(2), (2006)923-931.41.Shaher Momani,Solving a system of second-order obstacle problems a modified decompositionmethod,Applied Mathematics E-Notes,Vol.6,(2006)141-147.42.Shaher Momani and Zaid Odibat,Analytical approach to linear fractional partial differentialequations arising influid mechanics,Physics Letters A,Vol.355,(2006)271-279.43.Shaher Momani and Zaid Odibat,Analytical solution of a time-fractional Navier-Stokes equa-tion by Adomian decomposition method,Applied Mathematics and Computation,Vol.177,(2006) 488-494.44.Shaher Momani and Khaled Moadi,A reliable algorithm for solving fourth-order boundaryvalue problems,Journal of Applied Mathematics and Computing,Vol.22(3),(2006)185-197. 45.Ziad Odibat and Shaher Momani,Approximate solutions for boundary value problems oftime-fractional wave equation,Applied Mathematics and Computation,Vol.181(1),(2006)767-774.46.Shaher Momani and Rami Qaralleh,An efficient method for solving systems of fractionalintegro-differential equations,Computers and Mathematics with Application,Vol.52(3-4),(2006) 459-470.47.Ziad Odibat and Shaher Momani,Analytical spherically symmetric solution for the time-fractional Navier-Stokes equation,Advances in Theoretical and Apllied Mathematics(ATAM), Vol.1(2),(2006)97-107.48.Shaher Momani and Muhammad Aslam Noor,Numerical methods for fourth-order frac-tional integro-differential equations,Applied Mathematics and Computation,Vol.182(1),(2006) 754-760.949.Shaher Momani,A numerical scheme for the solution of multi-order fractional differential equa-tions,Applied Mathematics and Computation,Vol.182(1),(2006)761-770.50.Shaher Momani and Nabil Shawagfeh,Decomposition method for solving the fractionalRiccati differential equation,Applied Mathematics and Computation,Vol.182(2),(2006)1083-1092.51.Shaher Momani and Rami Qaralleh,Analytical approximate solution for a nonlinear frac-tional integro-differential equation,Nonlinear Analysis Forum Journal,Vol.11(2),(2006)237-249.52.Shaher Momani,Salah Abuasad and Zaid Odibat,Variational iteration method for solvingnon-linear boundary value problems,Applied Mathematics and Computation,Vol.183,(2006) 1351-1358.53.Shaher Momani,General solutions for the space-and time-fractional diffusion-wave equation,Journal of Physical Sciences,Vol.10,(2006)30-43.54.Shaher Momani and Zaid Odibat,Numerical comparison of methods for solving linear dif-ferential equations of fractional order,Chaos,Solitons&Fractals,Vol.31(5),(2007)1248-1255.55.Shaher Momani,An algorithm for solving a nonlinear fractional convection-diffusion problem,Communications in Nonlinear Science and Numerical Simulation,Vol.12(7),(2007)1283-1290.56.Shaher Momani and Zaid Odibat,Numerical approach to differential equations of fractionalorder,Journal of Computational and Applied Mathematics,Vol.207(1),(2007)96-110.57.Shaher Momani and Zaid Odibat,Fractional Green’s function for linear fractional inhomo-geneous partial differential equations influid mechanics,Journal of Applied Mathematics and Computing,Vol.24,(2007)167-178.58.Shaher Momani and Rami Qaralleh,Numerical approximations and Pade approximants fora fractional population growth model,Applied Mathematical Modelling,Vol.31(9),(2007)1907-1914.59.Shaher Momani and Ziad Odibat,Comparison between homotopy perturbation method andthe variational iteration method for linear fractional partial differential equations,Computers and Mathematics with Applications,Vol.54,(2007)910-919.60.Vedat Suat Erturk and Shaher Momani,Comparing numerical method for solving fourth-order boundary value problems,Applied Mathematics and Computation,Vol.188,(2007)1963-1968.61.Rabha Ibrahim and Shaher Momani,On the existence and uniqueness of solutions of aclass of fractional differential equations,Journal of Mathematical Analysis and Applications,Vol.334(1),(2007)1-10.62.Shaher Momani and Ziad Odibat,Homotopy perturbation method for nonlinear partial dif-ferential equations of fractional order,Physics Letters A,Vol.365,(2007)345-350.63.Zaid Odibat and Shaher Momani,A reliable treatment of homotopy perturbation methodfor Klein-Gordon equations,Physics Letters A,Vol.365,(2007)351-357.64.Khalida Inayat Noor,Muhammad Aslam Noor and Shaher Momani,Modified House-holder iterative method for nonlinear equations,Applied Mathematics and Computation,Vol.190, (2007)1534-1539.1065.Shaher Momani and Muhammad Aslam Noor,Numerical comparison of methods for solv-ing a special fourth-order boundary value problem,Applied Mathematics and Computation,Vol.191,(2007)218-224.66.Rabha Ibrahim and Shaher Momani,Multiple solutions for multi-order fractional differentialequations,Arab Journal of Mathematics and Mathematical Sciences,Vol.1,(2007)28-34.67.Shaher Momani,A decomposition method for solving unsteady convection-diffusion problems,Turkish Journal of Mathematics,Vol.31,(2007)1-10.68.Vedat Suat Erturk and Shaher Momani,A reliable algorithm for solving tenth-order bound-ary value problems,Numerical Algorithms,Vol.44(2),(2007)147-158.69.Zaid Odibat and Shaher Momani,Numerical solution of Fokker-Planck equation with space-and time-fractional derivtives,Physics Letters A,Vol.369(2007),349-358.70.Shaher Momani,Zaid Odibat and Vedat Suat Erturk,Generalized differential transformmethod for solving a space-and time-fractional diffusion-wave equation,Physics Letters A,Vol.370,(2007)379-387.71.Hossien Jafari and Shaher Momani,Solving fractional diffusion and wave equations by mod-ified homotopy perturbation method,Physics Letters A,Vol.370,(2007)388-396.72.Zaid Odibat and Shaher Momani,Numerical methods for nonlinear partial differential equa-tions of fractional order,Applied Mathematical modlling,Vol.32,(2008)28-39.73.Shaher Momani and Rabha Ibrahim,On a fractional integral equation of periodic func-tions involving Weyl-Riesz operator in Banach algebras,Journal of Mathematical Analysis and Applications,Vol.339,(2008)1210-1219.74.Zaid Odibat and Shaher Momani,Modified homotopy perturbation method:application toquadratic Riccati differential equation of fractional order,Chaos,Solitons&Fractals,Vol.36, (2008)167-174.75.Zaid Odibat and Shaher Momani,A generalized differential transform method for linearpartial differential equations of fractional order,Applied Mathematics Letters,Vol.21,(2008) 194-199.76.Shaher Momani and Ziad Odibat,A novel method for nonlinear fractional partial differentialequations:Combination of DTM and generalized Taylor’s formula,Journal of Computational and Applied Mathematics,Vol.220,(2008)85-95.77.G.H.Erjaee and Shaher Momani,Phase synchronization in fractional differential chaoticsystems.Physics Letters A,Vol.372,(2008)2350-2354.78.Vedat Suat Erturk Shaher Momani,and Zaid Odibat,Application of generalized dif-ferential transform method to multi-order fractional differential equations,Communications in Nonlinear Science and Numerical Simulation,Vol.13,(2008)1642-1654.79.Vedat Suat Erturk and Shaher Momani,Solving systems of fractional differential equationsusing differential transform method,Journal of Computational and Applied Mathematics,Vol.215,(2008)142-151.80.Shaher Momani and Vedat Suat Erturk,Solutions of non-linear oscillators by the modifieddifferential transform method,Computers&Mathematics with Applications,Vol.55,(2008) 833-842.1181.M.S.Chowdhury,Ishak Hashim and Shaher Momani,The multistage homotopy-perturbation method:A powerful scheme for handling the Lorenz system.Chaos,Solitons& Fractals,accepted(2007),doi:10.1016/j.chaos.2007.09.073.82.Omar Abdulaziz,Ishak Hashim and Shaher Momani,Application of homotopy-perturbation method to fractional IVPs,Journal of Computational and Applied Mathematics, Vol.216,(2008)574-584.83.Omar Abdulaziz,Ishak Hashim and Shaher Momani,Solving systems of fractional differ-ential equations by homotopy-perturbation method.Physics Letters A,Vol.372,(2008)541-549.84.Zaid Odibat,Shaher Momani and Vedat Suat Erturk,Generalized differential transformmethod:Application to differential equations of fractional order.Applied Mathematics and Com-putation,Vol.197,(2008)467-477.85.Omar Jaradat,Ahmad Al-Omari and Shaher Momani,Exsitence of the mild solutionfor fractional semilinear initial value problems,Nonlinear Analysis Journal:Theory,Methods& Applications,Vol.69,(2008)3153-3159.86.Ishak Hashim,Omar Abdulaziz and Shaher Momani,Homotopy analysis method for frac-tional IVPs,Communications in Nonlinear Science and Numerical Simulation,Vol.14,(2009) 674-784.87.Shaher Momani,Ziad Odibat and Ahmed Alawneh,Variational iteration method for solv-ing the space-and time-fractional KdV equation,Numerical Methods for Partial Differential Equations Journal,Vol.24(1),(2008)262-271.88.Shaher Momani and Vedat Suat Erturk,A numerical scheme for the solution of viscousCahn-Hilliard equation,Numerical Methods for Partial Differential Equations Journal,Vol.24(2), (2008)663-669.89.Shaher Momani and Ziad Odibat,Numerical solutions of the space-time fractional advection-dispersion equation,Numerical Methods for Partial Differential Equations Journal,accepted (2007),doi10.1002/num.20324.90.Zaid Odibat and Shaher Momani,Analytical comparison between the homotopy perturbationmethod and variational iteration method for differential equations of fractional order,International Journal of Modern Physics B,accepted(2007).91.Shaher Momani and Vedat Suat Erturk,Solving a system of fourth-order obstacle boundaryvalue problems by differential transform method,Kybernetes,Vol.73(2),(2008)315-325.92.Shaher Momani,Omar Jaradat and Rabha Ibrahim,Numerical approximations of a dy-namic system containing fractional derivatives.Journal of Applied Sciences,Vol.8(6),(2008) 1079-1084.93.Shaher Momani and Hossein Jafari,Numerical study of systems of fractional differentialequations by the decomposition method.Southeast Asian Bulletin of Mathematics,accepted (2006).94.S.M.Momani,Ahlam Jameel and Sora Al-Azwai,Local and global uniqueness theoremson fractional integro-differential equations via Gronwall’s and Bihari’s inequalities.Soochow Jour-nal of Mathematics,Vol.33(4),(2007)619-627.95.Vedat Suat Erturk and Shaher Momani,Differential transform technique for solvingfifth-order boundary value problems.Mathematical and Computational Applications,Vol.13(2), (2008)113-121.1296.Zaid Odibat and Shaher Momani,An algorithm for the numerical solution of differentialequations of fractional order,Journal of Applied Mathematics and Informatics,accepted(2007).97.Shaher Momani,Numerical simulations for the space-and time-fractional partial differentialequations,Proceedings of The Third Conferences On Research And Education In Mathematics, Malaysia,(2007)51-56.98.Vedat Suat Erturk and Shaher Momani,Differential transform method for obtaining positivesolutions for two-point nonlinear boundary value problems,International Journal:mathematical Manuscript(IJMM),Vol.1(1),(2007)65-72.99.Shaher Momani,Ziad Odibat and Ishak Hashim,Algorithms for nonlinear fractional par-tial differential equations:A selection of numerical methods,Topological Methods in Nonlinear Analysis Journal,Vol.31,(2008)211-226.100.Zaid Odibat,Shaher Momani and Ahmed Alawneh,Analytic study on time-fractional Schr¨o dinger equations:Exact solutions by GDTM,Journal of Physics:Conference Series,ac-cepted.101.Zaid Odibat,Shaher Momani,Applications of variational iteration and homotopy perturba-tion methods to fractional evolution equations,Topological Methods in Nonlinear Analysis Journal, Vol.31,(2008)227-234.102.Shaher Momani,An efficient numerical scheme for solving fractional convection-diffusion prob-lems.International Journal of Computational and Numerical Analysis and Applications,accepted (2007).103.Ziad Odibat,Shaher Momani and Ahmed Alawneh,Approximate analytical solution of the space-and time-fractional Burgers equations.Journal Europ´e en des Syst´e mes Automatis´e s, Vol.42(6),(2008)627-638.104.Shaher Momani and Rabha Ibrahim,Analytical solutions of a fractional oscillator by the decomposition method,International Journal of Pure and Applied Mathematics,Vol.37(1), (2007)119-132.105.M.H.Alnasr and Shaher Momani,Application of homotopy perturbation method to sin-gularly perturbed Volterra integral equations.Journal of Applied Sciences,Vol.8(6),(2008) 1073-1078.106.Ziad Odibat and Shaher Momani,Fractional Green’s Function for a Class of Fractional Partial Differential Equations.Journal Europ´e en des Syst´e mes Automatis´e s,Vol.42(6),(2008) 639-652.107.Shaher Momani,Vedat Suat Erturk and Sana Abu-Qurra,An approximation of the analytic solution of the Helmholtz equation.International Journal:mathematical Manuscript (IJMM),accepted.108.Vedat Suat Erturk and Shaher Momani,solutions of two forms of Blasius equation on a half-infinite domain.Journal of Algorithms and Computational Technology,accepted.109.Rabha Ibrahim and Shaher Momani,Upper and lower bounds of solutions for fractional integral equations.Surveys in Mathematics and its Applications,Vol.24,(2007)145-156. 110.Shaher Momani,G.H.Erjaee and M.H.Alnasr,The modified homotopy perturbation method for handling non-linear puters and Mathematics with Applications,ac-cepted.13。

观测噪声方差变化条件下系统状态估计方法

观测噪声方差变化条件下系统状态估计方法

观测噪声方差变化条件下系统状态估计方法宋婉娟; 张剑【期刊名称】《《探测与控制学报》》【年(卷),期】2019(041)005【总页数】6页(P96-100,105)【关键词】非线性状态估计; 容积卡尔曼滤波; 变分贝叶斯滤波; 噪声统计特性【作者】宋婉娟; 张剑【作者单位】湖北第二师范学院计算机学院湖北武汉 430205; 武汉体育学院体育工程与信息技术学院湖北武汉 430070【正文语种】中文【中图分类】TP202; TP3910 引言非线性系统状态估计问题是工程应用中的难点和瓶颈,备受研究人员关注[1-3],但是由于该类系统参量的未知性,很难建立精确的数学模型,目前常用的解决方法主要是采用滤波估计,数值逼近的思路解决。

例如,扩展卡尔曼滤波(Extended Kalman Filter,EKF)[4-7],通过对非线性系统进行一阶泰勒近似,将非线性系统进行局部线性近似处理,虽简单易行,但存在明显的截断误差,噪声敏感性强;无迹卡尔曼滤波(Unscented Kalman Filter, UKF)[8-10],通过对模型统计参量的非线性变换,实现系统的高阶近似,但是其估计精度严重依赖于系统初值和观测噪声,当噪声参量时变情况下跟踪精度会降低;粒子滤波(Particle Filter, PF)[11-15]通过粒子加权求和的方法对系统积分进行拟合,在粒子数足够多的情况下是可以无限逼近系统真实状态的,但是该方法的滤波精度和实时性之间的矛盾至今尚未有效解决,限制了其工程应用。

针对这些问题,文献[16]采用高阶球面积分传播容积点的思想提出了容积卡尔曼滤波(Cubature Kalman Filter, CKF)方法,对状态的均值和协方差进行非线性传播,同EKF、UKF以及PF等非线性滤波方法相比,不仅提升了滤波精度,计算的复杂性也大大降低,在目标跟踪、状态估计等领域得到了广泛的应用。

但是标准CKF方法严重依赖于精确的状态模型和噪声统计特性[17-19],而在实际的工程应用中很难满足,特别是观测噪声方差,具有较强的时变特性,很难满足噪声方差统计特性精确已知的要求[20],限制了其在工程中的推广应用。

AI专用词汇

AI专用词汇

AI专⽤词汇LetterAAccumulatederrorbackpropagation累积误差逆传播ActivationFunction激活函数AdaptiveResonanceTheory/ART⾃适应谐振理论Addictivemodel加性学习Adversari alNetworks对抗⽹络AffineLayer仿射层Affinitymatrix亲和矩阵Agent代理/智能体Algorithm算法Alpha-betapruningα-β剪枝Anomalydetection异常检测Approximation近似AreaUnderROCCurve/AUCRoc曲线下⾯积ArtificialGeneralIntelligence/AGI通⽤⼈⼯智能ArtificialIntelligence/AI⼈⼯智能Associationanalysis关联分析Attentionmechanism注意⼒机制Attributeconditionalindependenceassumption属性条件独⽴性假设Attributespace属性空间Attributevalue属性值Autoencoder⾃编码器Automaticspeechrecognition⾃动语⾳识别Automaticsummarization⾃动摘要Aver agegradient平均梯度Average-Pooling平均池化LetterBBackpropagationThroughTime通过时间的反向传播Backpropagation/BP反向传播Baselearner基学习器Baselearnin galgorithm基学习算法BatchNormalization/BN批量归⼀化Bayesdecisionrule贝叶斯判定准则BayesModelAveraging/BMA贝叶斯模型平均Bayesoptimalclassifier贝叶斯最优分类器Bayesiandecisiontheory贝叶斯决策论Bayesiannetwork贝叶斯⽹络Between-cla ssscattermatrix类间散度矩阵Bias偏置/偏差Bias-variancedecomposition偏差-⽅差分解Bias-VarianceDilemma偏差–⽅差困境Bi-directionalLong-ShortTermMemory/Bi-LSTM双向长短期记忆Binaryclassification⼆分类Binomialtest⼆项检验Bi-partition⼆分法Boltzmannmachine玻尔兹曼机Bootstrapsampling⾃助采样法/可重复采样/有放回采样Bootstrapping⾃助法Break-EventPoint/BEP平衡点LetterCCalibration校准Cascade-Correlation级联相关Categoricalattribute离散属性Class-conditionalprobability类条件概率Classificationandregressiontree/CART分类与回归树Classifier分类器Class-imbalance类别不平衡Closed-form闭式Cluster簇/类/集群Clusteranalysis聚类分析Clustering聚类Clusteringensemble聚类集成Co-adapting共适应Codin gmatrix编码矩阵COLT国际学习理论会议Committee-basedlearning基于委员会的学习Competiti velearning竞争型学习Componentlearner组件学习器Comprehensibility可解释性Comput ationCost计算成本ComputationalLinguistics计算语⾔学Computervision计算机视觉C onceptdrift概念漂移ConceptLearningSystem/CLS概念学习系统Conditionalentropy条件熵Conditionalmutualinformation条件互信息ConditionalProbabilityTable/CPT条件概率表Conditionalrandomfield/CRF条件随机场Conditionalrisk条件风险Confidence置信度Confusionmatrix混淆矩阵Connectionweight连接权Connectionism连结主义Consistency⼀致性/相合性Contingencytable列联表Continuousattribute连续属性Convergence收敛Conversationalagent会话智能体Convexquadraticprogramming凸⼆次规划Convexity凸性Convolutionalneuralnetwork/CNN卷积神经⽹络Co-oc currence同现Correlationcoefficient相关系数Cosinesimilarity余弦相似度Costcurve成本曲线CostFunction成本函数Costmatrix成本矩阵Cost-sensitive成本敏感Crosse ntropy交叉熵Crossvalidation交叉验证Crowdsourcing众包Curseofdimensionality维数灾难Cutpoint截断点Cuttingplanealgorithm割平⾯法LetterDDatamining数据挖掘Dataset数据集DecisionBoundary决策边界Decisionstump决策树桩Decisiontree决策树/判定树Deduction演绎DeepBeliefNetwork深度信念⽹络DeepConvolutionalGe nerativeAdversarialNetwork/DCGAN深度卷积⽣成对抗⽹络Deeplearning深度学习Deep neuralnetwork/DNN深度神经⽹络DeepQ-Learning深度Q学习DeepQ-Network深度Q⽹络Densityestimation密度估计Density-basedclustering密度聚类Differentiab leneuralcomputer可微分神经计算机Dimensionalityreductionalgorithm降维算法D irectededge有向边Disagreementmeasure不合度量Discriminativemodel判别模型Di scriminator判别器Distancemeasure距离度量Distancemetriclearning距离度量学习D istribution分布Divergence散度Diversitymeasure多样性度量/差异性度量Domainadaption领域⾃适应Downsampling下采样D-separation(Directedseparation)有向分离Dual problem对偶问题Dummynode哑结点DynamicFusion动态融合Dynamicprogramming动态规划LetterEEigenvaluedecomposition特征值分解Embedding嵌⼊Emotionalanalysis情绪分析Empiricalconditionalentropy经验条件熵Empiricalentropy经验熵Empiricalerror经验误差Empiricalrisk经验风险End-to-End端到端Energy-basedmodel基于能量的模型Ensemblelearning集成学习Ensemblepruning集成修剪ErrorCorrectingOu tputCodes/ECOC纠错输出码Errorrate错误率Error-ambiguitydecomposition误差-分歧分解Euclideandistance欧⽒距离Evolutionarycomputation演化计算Expectation-Maximization期望最⼤化Expectedloss期望损失ExplodingGradientProblem梯度爆炸问题Exponentiallossfunction指数损失函数ExtremeLearningMachine/ELM超限学习机LetterFFactorization因⼦分解Falsenegative假负类Falsepositive假正类False PositiveRate/FPR假正例率Featureengineering特征⼯程Featureselection特征选择Featurevector特征向量FeaturedLearning特征学习FeedforwardNeuralNetworks/FNN前馈神经⽹络Fine-tuning微调Flippingoutput翻转法Fluctuation震荡Forwards tagewisealgorithm前向分步算法Frequentist频率主义学派Full-rankmatrix满秩矩阵Func tionalneuron功能神经元LetterGGainratio增益率Gametheory博弈论Gaussianker nelfunction⾼斯核函数GaussianMixtureModel⾼斯混合模型GeneralProblemSolving通⽤问题求解Generalization泛化Generalizationerror泛化误差Generalizatione rrorbound泛化误差上界GeneralizedLagrangefunction⼴义拉格朗⽇函数Generalized linearmodel⼴义线性模型GeneralizedRayleighquotient⼴义瑞利商GenerativeAd versarialNetworks/GAN⽣成对抗⽹络GenerativeModel⽣成模型Generator⽣成器Genet icAlgorithm/GA遗传算法Gibbssampling吉布斯采样Giniindex基尼指数Globalminimum全局最⼩GlobalOptimization全局优化Gradientboosting梯度提升GradientDescent梯度下降Graphtheory图论Ground-truth真相/真实LetterHHardmargin硬间隔Hardvoting硬投票Harmonicmean调和平均Hessematrix海塞矩阵Hiddendynamicmodel隐动态模型H iddenlayer隐藏层HiddenMarkovModel/HMM隐马尔可夫模型Hierarchicalclustering层次聚类Hilbertspace希尔伯特空间Hingelossfunction合页损失函数Hold-out留出法Homo geneous同质Hybridcomputing混合计算Hyperparameter超参数Hypothesis假设Hypothe sistest假设验证LetterIICML国际机器学习会议Improvediterativescaling/IIS改进的迭代尺度法Incrementallearning增量学习Independentandidenticallydistributed/i.i.d.独⽴同分布IndependentComponentAnalysis/ICA独⽴成分分析Indicatorfunction指⽰函数Individuallearner个体学习器Induction归纳Inductivebias归纳偏好I nductivelearning归纳学习InductiveLogicProgramming/ILP归纳逻辑程序设计Infor mationentropy信息熵Informationgain信息增益Inputlayer输⼊层Insensitiveloss不敏感损失Inter-clustersimilarity簇间相似度InternationalConferencefor MachineLearning/ICML国际机器学习⼤会Intra-clustersimilarity簇内相似度Intrinsicvalue固有值IsometricMapping/Isomap等度量映射Isotonicregression等分回归It erativeDichotomiser迭代⼆分器LetterKKernelmethod核⽅法Kerneltrick核技巧K ernelizedLinearDiscriminantAnalysis/KLDA核线性判别分析K-foldcrossvalidationk折交叉验证/k倍交叉验证K-MeansClusteringK–均值聚类K-NearestNeighb oursAlgorithm/KNNK近邻算法Knowledgebase知识库KnowledgeRepresentation知识表征LetterLLabelspace标记空间Lagrangeduality拉格朗⽇对偶性Lagrangemultiplier拉格朗⽇乘⼦Laplacesmoothing拉普拉斯平滑Laplaciancorrection拉普拉斯修正Latent DirichletAllocation隐狄利克雷分布Latentsemanticanalysis潜在语义分析Latentvariable隐变量Lazylearning懒惰学习Learner学习器Learningbyanalogy类⽐学习Learn ingrate学习率LearningVectorQuantization/LVQ学习向量量化Leastsquaresre gressiontree最⼩⼆乘回归树Leave-One-Out/LOO留⼀法linearchainconditional randomfield线性链条件随机场LinearDiscriminantAnalysis/LDA线性判别分析Linearmodel线性模型LinearRegression线性回归Linkfunction联系函数LocalMarkovproperty局部马尔可夫性Localminimum局部最⼩Loglikelihood对数似然Logodds/logit对数⼏率Lo gisticRegressionLogistic回归Log-likelihood对数似然Log-linearregression对数线性回归Long-ShortTermMemory/LSTM长短期记忆Lossfunction损失函数LetterM Machinetranslation/MT机器翻译Macron-P宏查准率Macron-R宏查全率Majorityvoting绝对多数投票法Manifoldassumption流形假设Manifoldlearning流形学习Margintheory间隔理论Marginaldistribution边际分布Marginalindependence边际独⽴性Marginalization边际化MarkovChainMonteCarlo/MCMC马尔可夫链蒙特卡罗⽅法MarkovRandomField马尔可夫随机场Maximalclique最⼤团MaximumLikelihoodEstimation/MLE极⼤似然估计/极⼤似然法Maximummargin最⼤间隔Maximumweightedspanningtree最⼤带权⽣成树Max-P ooling最⼤池化Meansquarederror均⽅误差Meta-learner元学习器Metriclearning度量学习Micro-P微查准率Micro-R微查全率MinimalDescriptionLength/MDL最⼩描述长度Minim axgame极⼩极⼤博弈Misclassificationcost误分类成本Mixtureofexperts混合专家Momentum动量Moralgraph道德图/端正图Multi-classclassification多分类Multi-docum entsummarization多⽂档摘要Multi-layerfeedforwardneuralnetworks多层前馈神经⽹络MultilayerPerceptron/MLP多层感知器Multimodallearning多模态学习Multipl eDimensionalScaling多维缩放Multiplelinearregression多元线性回归Multi-re sponseLinearRegression/MLR多响应线性回归Mutualinformation互信息LetterN Naivebayes朴素贝叶斯NaiveBayesClassifier朴素贝叶斯分类器Namedentityrecognition命名实体识别Nashequilibrium纳什均衡Naturallanguagegeneration/NLG⾃然语⾔⽣成Naturallanguageprocessing⾃然语⾔处理Negativeclass负类Negativecorrelation负相关法NegativeLogLikelihood负对数似然NeighbourhoodComponentAnalysis/NCA近邻成分分析NeuralMachineTranslation神经机器翻译NeuralTuringMachine神经图灵机Newtonmethod⽜顿法NIPS国际神经信息处理系统会议NoFreeLunchTheorem /NFL没有免费的午餐定理Noise-contrastiveestimation噪⾳对⽐估计Nominalattribute列名属性Non-convexoptimization⾮凸优化Nonlinearmodel⾮线性模型Non-metricdistance⾮度量距离Non-negativematrixfactorization⾮负矩阵分解Non-ordinalattribute⽆序属性Non-SaturatingGame⾮饱和博弈Norm范数Normalization归⼀化Nuclearnorm核范数Numericalattribute数值属性LetterOObjectivefunction⽬标函数Obliquedecisiontree斜决策树Occam’srazor奥卡姆剃⼑Odds⼏率Off-Policy离策略Oneshotlearning⼀次性学习One-DependentEstimator/ODE独依赖估计On-Policy在策略Ordinalattribute有序属性Out-of-bagestimate包外估计Outputlayer输出层Outputsmearing输出调制法Overfitting过拟合/过配Oversampling过采样LetterPPairedt-test成对t检验Pairwise成对型PairwiseMarkovproperty成对马尔可夫性Parameter参数Parameterestimation参数估计Parametertuning调参Parsetree解析树ParticleSwarmOptimization/PSO粒⼦群优化算法Part-of-speechtagging词性标注Perceptron感知机Performanceme asure性能度量PlugandPlayGenerativeNetwork即插即⽤⽣成⽹络Pluralityvoting相对多数投票法Polaritydetection极性检测Polynomialkernelfunction多项式核函数Pooling池化Positiveclass正类Positivedefinitematrix正定矩阵Post-hoctest后续检验Post-pruning后剪枝potentialfunction势函数Precision查准率/准确率Prepruning预剪枝Principalcomponentanalysis/PCA主成分分析Principleofmultipleexplanations多释原则Prior先验ProbabilityGraphicalModel概率图模型ProximalGradientDescent/PGD近端梯度下降Pruning剪枝Pseudo-label伪标记LetterQQuantizedNeu ralNetwork量⼦化神经⽹络Quantumcomputer量⼦计算机QuantumComputing量⼦计算Quasi Newtonmethod拟⽜顿法LetterRRadialBasisFunction/RBF径向基函数RandomFo restAlgorithm随机森林算法Randomwalk随机漫步Recall查全率/召回率ReceiverOperatin gCharacteristic/ROC受试者⼯作特征RectifiedLinearUnit/ReLU线性修正单元Recurr entNeuralNetwork循环神经⽹络Recursiveneuralnetwork递归神经⽹络Referencemodel参考模型Regression回归Regularization正则化Reinforcementlearning/RL强化学习Representationlearning表征学习Representertheorem表⽰定理reproducingke rnelHilbertspace/RKHS再⽣核希尔伯特空间Re-sampling重采样法Rescaling再缩放Residu alMapping残差映射ResidualNetwork残差⽹络RestrictedBoltzmannMachine/RBM受限玻尔兹曼机RestrictedIsometryProperty/RIP限定等距性Re-weighting重赋权法Robu stness稳健性/鲁棒性Rootnode根结点RuleEngine规则引擎Rulelearning规则学习LetterS Saddlepoint鞍点Samplespace样本空间Sampling采样Scorefunction评分函数Self-Driving⾃动驾驶Self-OrganizingMap/SOM⾃组织映射Semi-naiveBayesclassifiers半朴素贝叶斯分类器Semi-SupervisedLearning半监督学习semi-SupervisedSupportVec torMachine半监督⽀持向量机Sentimentanalysis情感分析Separatinghyperplane分离超平⾯SigmoidfunctionSigmoid函数Similaritymeasure相似度度量Simulatedannealing模拟退⽕Simultaneouslocalizationandmapping同步定位与地图构建SingularV alueDecomposition奇异值分解Slackvariables松弛变量Smoothing平滑Softmargin软间隔Softmarginmaximization软间隔最⼤化Softvoting软投票Sparserepresentation稀疏表征Sparsity稀疏性Specialization特化SpectralClustering谱聚类SpeechRecognition语⾳识别Splittingvariable切分变量Squashingfunction挤压函数Stability-plasticitydilemma可塑性-稳定性困境Statisticallearning统计学习Statusfeaturefunction状态特征函Stochasticgradientdescent随机梯度下降Stratifiedsampling分层采样Structuralrisk结构风险Structuralriskminimization/SRM结构风险最⼩化S ubspace⼦空间Supervisedlearning监督学习/有导师学习supportvectorexpansion⽀持向量展式SupportVectorMachine/SVM⽀持向量机Surrogatloss替代损失Surrogatefunction替代函数Symboliclearning符号学习Symbolism符号主义Synset同义词集LetterTT-Di stributionStochasticNeighbourEmbedding/t-SNET–分布随机近邻嵌⼊Tensor张量TensorProcessingUnits/TPU张量处理单元Theleastsquaremethod最⼩⼆乘法Th reshold阈值Thresholdlogicunit阈值逻辑单元Threshold-moving阈值移动TimeStep时间步骤Tokenization标记化Trainingerror训练误差Traininginstance训练⽰例/训练例Tran sductivelearning直推学习Transferlearning迁移学习Treebank树库Tria-by-error试错法Truenegative真负类Truepositive真正类TruePositiveRate/TPR真正例率TuringMachine图灵机Twice-learning⼆次学习LetterUUnderfitting⽋拟合/⽋配Undersampling⽋采样Understandability可理解性Unequalcost⾮均等代价Unit-stepfunction单位阶跃函数Univariatedecisiontree单变量决策树Unsupervisedlearning⽆监督学习/⽆导师学习Unsupervisedlayer-wisetraining⽆监督逐层训练Upsampling上采样LetterVVanishingGradientProblem梯度消失问题Variationalinference变分推断VCTheoryVC维理论Versionspace版本空间Viterbialgorithm维特⽐算法VonNeumannarchitecture冯·诺伊曼架构LetterWWassersteinGAN/WGANWasserstein⽣成对抗⽹络Weaklearner弱学习器Weight权重Weightsharing权共享Weightedvoting加权投票法Within-classscattermatrix类内散度矩阵Wordembedding词嵌⼊Wordsensedisambiguation词义消歧LetterZZero-datalearning零数据学习Zero-shotlearning零次学习。

gaussian 教程

说明........................................................................................................................................... 1 前言............................................................................................................................................. 1 运行 Gaussian ............................................................................................................................. 2
Exploring Chemistry with Electronic Structure Method
Sesman Aeleen Frisch
Gaussian, Inc Pittsburgh, PA 2002 年 9 月 25 日
特别声明
本文转自南开大学 BBS 网站,在此对译者表示衷心感谢!!!!
Unix/Linux 平台 .................................................................................................................. 2 Windows 平台.................................................................

Gaussian简介与案例

笛卡尔坐标或内坐标
6.Gaussian的示例
• 步骤四:查看单点能计算结果
E表示计算能量值 迭带了6次 Convg收敛标准, -V/T 位力定理 值为2表示结果合理 S**2表示自旋污染
6.Gaussian的示例
• 步骤五:优化结构,调整输入文件分子几何构型的变 化对能量有很大的影响在计算执行路径加关键词opt
热力学参数
基态和激发态能量 化学键的键能 •能量计算 电子亲合能和电离能 化学反应途径和势能面 •光谱计算
IR光谱 Raman光谱 吸收/发射光谱 NMR
适用体系:气相和溶液
4.Gaussian 09程序架构

Gaussian程序在设计上实现了模块化,即不同功能的 程序设计在不同的模块中;
模块名称 Link 0 Link 1 Link 1xx Link 202 功能 初始化程序并控制模块; 处理Route部分、根据关键词创建需要执行Links列表、 初始化临时文件; 读入题目和分子信息、各种优化算法、动力学计算相 关Link、BSSE能量整合相关Links; 坐标、结构变量和对称性检查、坐标标准化;
9
6.Gaussian的示例
• 步骤一: 用Gaussian View作一Байду номын сангаас醛分子,保存后查看
6.Gaussian的示例
• 步骤二:将甲醛分子的输入文件导入Gaussian
6.Gaussian的示例
• 步骤三:设定Gaussian运行参数
计算机信息 理论等级,基组 题目
体系总电荷和多重度 分子坐标
GAUSSIAN 92,
GAUSSIAN 94, GAUSSIAN 98, GAUSSIAN 03,
GAUSSIAN 09

Graduated Nonconvexity by Functional Focusing

Graduated Nonconvexity by Functional FocusingMads NielsenAbstract —Reconstruction of noise-corrupted surfaces may be stated as a (in general nonconvex) functional minimization problem. For functionals with quadratic data term, this paper addresses the criteria for such functionals to be convex, and the variational approach for minimization. I present two automatic and general methods ofapproximation with convex functionals based on Gaussian convolution.They are compared to the Blake-Zisserman graduated nonconvexity (GNC) method and Bilbro et al. and Geiger and Girosi’s mean field annealing (MFA) of a weak membrane.Index Terms —Graduated nonconvexity, functional minimization, mean field annealing, Bayesian reconstruction.———————— 3 ————————1I NTRODUCTIONT HE reconstruction of noise-corrupted surfaces cannot in general be deduced . Information is lost, and the reconstruction must be inferred by an estimation paradigm such as Bayesian Maximum A Posteriori Estimation (MAP) or Minimum Description Length (MDL) [1]. MAP selects the most probable reconstruction while MDL selects the simplest explanation of the measurements. Under appropriate conditions [2], the two methodologies yield identical formulations, and I use the Bayesian formulation.A sampled surface D n :ޚޒa is noise-corrupted by additive noise N n :ޚޒa yielding the measurementsM (x ) = D (x ) + N (x ).The MAP estimate is found as the argument R n:ޚޒa maxi-mizing the a posteriori probabilityp R D M p M D R p D R p M ====ch c h b g b gwhere p (M ) can be perceived as a normalizing constant. This is maximized by the same argument R that minimizesE R p R D M cp M D R p D R E R E Rd s ∫-=+=-=-=∫+log log log c hc h b g(1)where E d is the data term and E s is the smoothness term . Assuming a model of independent identically distributed Gaussian noise, the data term becomes quadratic (Also spatially correlated and non-stationary Gaussian noise leads to a quadratic term as long as it is not correlated with the image.) in R , leading to (the inverse noise variance l s 22∫- is absorbed in E s [R ])E R MRE R ii s i =-+ŒÂe j2W(2)where superscript i denotes the i th point in the discrete surfacedomain W . In general, the smoothness term can express any prior knowledge on the surface, but here I assume the prior to be inde-pendent identically distributed in the gradient —R . That is,E R f R p R s ii i i=—∫-—ŒŒÂÂe j e j l 2WWlog (3)where f n :ޒޒa is the smoothness function . In the following, the convexity of the functional E [R ] is analyzed for various smooth-ness functions f .The shape of f determines the properties of the optimal recon-struction R . If f is quadratic, the minimization implies standard regularization as formulated by Tikhonov and Arsenin [3]. In this case, the functional E is convex and the unique solution can be found by a linear nonrecursive or recursive filtering [4], [5].Blake and Zisserman reformulated (by use of what Rangarajan and Chelleppa [6] call the adiabatic approximation) the discon-tinuous surface reconstruction problem of Geman and Geman [7]on the form of (3) so thatf TT x x x b g =<R S |T|l l 222222if otherwise(4)where ʈ◊ʈ indicates the Euclidean norm. In this case the solution is called the weak membrane, and the functional E is nonconvex and leads to a serious minimization problem. Other smoothness func-tions f have been used in computer vision [8]. The assumption of 3D isotropy of surface normals implies the Lorentzian estimator f x xb gej =+log 12[9] while the assumption of surface structurelike Brownian motion leads to a more complicated form of f [10].In the following, I analyze the class of smoothness functions f im-plying convex functionals E . This analysis is followed by the de-velopment and characterization of variational methods for ap-proximation of the solution in the nonconvex case. A comparison with and analysis of the Blake-Zisserman GNC and the MFA of the weak string is performed.2C ONVEXITY OF F UNCTIONALSThe convexity of a functional on the form of (2) and (3) can be specified directly as a constraint on the smoothness function f :T HEOREM 1 [1D C ONVEXITY ]. E [R ] on the form of (2) is convex if"Œ¢¢>-x G f x :b g12, where G is the set of possible gradient val-ues —R .P ROOF . Let N = |W |. E [R ] is convex if the Hessian H of E withrespect to R is positive definite: "Œ>x x Hx ޒNT:0. The Hessian can be written H = 2I + F , whereF f f f f f f f f f f f f f f N N N NNNN=¢¢-¢¢-¢¢¢¢+¢¢-¢¢-¢¢◊◊◊◊-¢¢-¢¢¢¢+¢¢-¢¢-¢¢¢¢L N M M M M M M O QPP P P PP ---222233111¢¢∫¢¢—f f R iic h , and the operator —iis defined as —∫--i i i x x x 1. Since "Œ≥—=Âx x x ޒN ii N :2222c h , the cri-terion of H being positive definite"+—¢¢>=Âx xx i:20222c h i Nif is always true if "Œ¢¢>-i N f i 212K :. This is true if ¢¢>-f x b g12 for every possible gradient value x Œ G .0162-8828/97/$10.00 © 1997 IEEE————————————————• The author is with 3D-Lab, School of Dentistry, Nørre Allé 25, DK-2200Copenhagen N, Denmark. E-mail: malte@lab3d.odont.ku.dk.Manuscript received Nov. 28, 1994; revised Feb. 24, 1997. Recommended for acceptance by J. Malik.For information on obtaining reprints of this article, please send e-mail to:transpami@, and reference IEEECS Log Number 104687.Normally I will identify G = ޒ, but some discussions regarding MFA simplifies using distinct notation. Notice, in the admissible class of smoothness functions f implying a convex functional the second derivative may indeed be negative and introduce disconti-nuities in the reconstruction. The above theorem is valid for inte-ger sampled signals in 1D where f is a function of the gradient.When another sampling h is used and f is a function of a derivative of order m the theorem still holds but with a modified limit on ¢¢f x b g[11]:¢¢>-+--F H I K--f x h m m m mb g211211In the case where the measurements M to be reconstructed are of ahigher dimensionality M n:ޒޒa , the picture complicates a bit.T HEOREM 2 [ND C ONVEXITY ]. A functional E [R ], R n :ޒޒa onthe form of (2) is convex if the Hessian H of the smoothness func-tion f is in the class +=Œ"Œ£>-R S TU V W¥•H y yy Hy ޒޒn nn T {}:112where ◊•denotes the maximum norm .The proof is analogous to the above [11]. Due to the square gridon which R is defined, the criterion is not rotationally symmetric.An isotropic separation reads: If the eigenvalues l i of the Hessian of f (x ) are all larger than -1/2n for every x , then E is convex. If one of the eigenvalues is smaller than -1/2, then E is in general non-convex. Notice that the different dimensions cannot help each other to gain convexity. On the contrary, they all have to share the convexity provided by the data term.3G RADUATED N ONCONVEXITYWhen the functional to be minimized is nonconvex, gradient methods do not guarantee obtaining the global minimum.Variational methods estimate the solution using an embedding of the functional E [R ] into a one-parameter family of functionals E s [R ] where E 0[R ] ∫ E [R ]. The (local) minimum of E s [R ] is tracked when the control parameter s is varied from s 0 to 0. The final estimate is a local minimum of E [R ] but not necessarily the global minimum. In order to argue about the quality of a varia-tional algorithm, I describe three properties: the initial functional E s 0, the general evolution of the functional, and the convergence of the functional.In a graduated nonconvexity (GNC) algorithm, the initial func-tional E R s 0 must be convex. Thus the estimate becomes inde-pendent of the initial state and may then be uniquely defined. A set of lines of steepest descent connects to a minimum and forms its basin of attraction. If the functional is convex, every function R is in the basin of attraction of the global minimum. If the initial functional E R s 0 contains several minima, the initial estimate will generically be contained in one basin defining the initially chosen minimum. Thus the final estimate may depend on the initial con-dition. Obtaining convexity in the initial functional is a critical point of a variational algorithm.The evolution of the functional and especially the creation and displacement of nonconvexities in the functional is vital to the final choice of minimum. In general, minima will be created and annihilated when varying the control parameter of the functional. The creation (or annihilation) happens through the catastrophes classified by Thom [12]. During the simplest event,the fold , minima appear/disappear on hillsides and the solution is still uniquely defined even though it may move discontinu-ously as a function of the control parameter [13]. During a cusp ,however, a minimum creation corresponds to a balanced split-ting of the minimum in two, and a unique tracking of the solu-tion is not defined. That is, cusps should not happen generically in the functional. The analysis of catastrophes in the functional is, even though important, not the subject of this paper.The displacement of the minimum during variation is impor-tant for the final solution. It is hard generally to quantify this dis-placement, but in the case of MFA of the weak membrane [14], [15]one can describe parts of the dynamics of the local minima and thereby derive some qualitative structure of the solution.Finally, the local minima of E s must converge to the local minima of E . This is captured in the mathematical concept of G -convergence [16]. In general, functions can be G -convergent with-out being uniform convergent and vice versa, but in a discrete setting uniform convergence implies G -convergence. Hence, I am in this context satisfied by uniformly convergence for GNC-algorithms.The GNC of the weak string by Blake and Zisserman [17] fulfils the above criteria of convexity and convergence. Here, the critical part of f is substituted by a negative parabola of second derivative larger than -1/2 and the interval of substitution shrinks with s . In the following, two GNC generating algorithms are presented.They work on a large class of smoothness functions and are auto-matic. Furthermore, the MFA of the weak string is reviewed in terms of the above properties of variational algorithms.4F OCUSING A LGORITHMSTwo GNC algorithms based on Gaussian blurring of the function-als are presented. The first, Smoothness Focusing (SF), performs a Gaussian blurring of the smoothness function f while the second,Probability Focusing (PF), performs a Gaussian blurring of the prior probability p x ef xb gb g ∫-.T HEOREM 3 [S MOOTHNESS B LURRING ]. Any functional E on theform of (2) having a smoothness function f (x ) ∫ g (x ) + h (x ), where h (x ) is integrable and g implies a convex functional E *= E d + Âg ,becomes convex by convolution of f (x ) with a Gaussian of appro-priate standard deviation s 0.P ROOF . From the convexity of E * and Theorem 1 follows "x Œޒ: g ¢¢(x ) ≥ b , where b >-12. Furthermore,Gf x dkG kg x k dkG x kh kbdkG k dk h k G x k b e dk h kx ss s ssspଙc h b g b g b g b g b gb g b g b gb g ≤=¢¢-+¢¢-≥-¢¢-=-z zzzzŒ-ޒޒޒޒޒޒsup /22323This implies that "Œ≤>-x G f x ޒ:/s 012ଙej b gifs p3322122>+-zeb dk h k//b gb gޒand thereby (Theorem 1) E s 0 is convex.In the case of the weak string (the 1D version of the weak membrane (4)), f (x ) can be constructed fromg x T h x x Tx Tb gb g ==-<R S Tl l l 22222222and if otherwise implying b = 0 anddk h k T ޒz=b g 4323l /. According to Theorem 3,the weak string becomes convex by a convolution of f with a Gaus-sian ofs 0 > 2e-1/2(9p /2)-1/6l 2/3T Ϸ 0.695l2/3TThis is a conservative estimate of the lower bound on s 0 and in reality approximately 9 percent higher than necessary. Notice, also functionals not mentioned in Theorem 3 may be convex after smoothness blurring. Examples are periodic smoothness functions which all imply convex approximations [11].T HEOREM 4 [P ROBABILITY B LURRING ]. The functional of (2) be-comes convex by a convolution of the prior distribution p x ef xb gb g ∫- by a Gaussian of finite standard deviation s > s 0[p ]when p is Gaussian except in a finite interval .The proof is given in [18]. It is also valid if the Gaussian part ofp has an infinite standard deviation, and thereby the Probability Blurring applies to the weak string.Both Smoothness and Probability Blurring have here been de-fined for the 1D case, but generalize to higher dimensions since Gaussian convolution is separable [11].These two blurring methods generating convex approximations of nonconvex functionals can be used to design GNC algorithms,here called focusing algorithms : The global minimum is found for s = s 0 and then locally tracked as s is lowered towards zero. In Fig. 1 the Smoothness Focusing behavior is compared to the Blake-Zisserman GNC. The SF finds a solution of lower energy for most image gradients. In general the BZ finds too many discontinuities while the SF finds too few. In Fig. 2 the development of disconti-nuities during the Smoothness Focusing of the weak membrane on an image of a water lily is shown.The focusing algorithms can be given a Bayesian rationale as respectively maximizing the expectation value of the a posterior probability and minimizing the expectation value of the energy functionals when the gradient can only be measured with noise from the discretely sampled image [19].In Fig. 3 the Probability Focusing is applied to the Lorentzian estimator f (x ) = l 2log(1 + |x ʈ2) assuming 3D isotropy of surface normals [9].Fig. 1. The energy of the weak string (l = 1, T = 1) as a function of the gradient of the input signal of the solutions found by Blake-Zisserman GNC and Smoothness Focusing. The BZ detects some discontinuitiesfor too small gradients while the SF detects discontinuities a little too late when the gradient is increased.Fig. 2. Development of discontinuities during Smoothness Focusing of weak membrane (l = 25, T = 3). From upper left: original image, gradi-ents larger than T indicated for s = 15, 12.5, 10.42, 6.03, 2.91, 1.17,0.47, 0.19, and the resulting image.Fig. 3. Reconstruction using the Lorentzian estimator and Probability Focusing. From upper left: original data, noise corrupted signal, width SNR =2 on the step edges, the reconstruction, and the normalized residual.5M EAN F IELD A NNEALINGMean field annealing (MFA) is a deterministic analogy of simu-lated annealing. Instead of randomly sampling a Gibbs distribu-tion of solutions, the mean of the distribution is computed:R dRRp R dRReZE R ∫∫zz-b gswhere Z , the partition function, normalises the distribution.When s tends towards infinity, all solutions have same prob-ability and the MF approximation becomes the center of gravity of the domain of R . When s tends to zero, only the solution of minimum energy has a nonzero probability, and the minimum-energy-solution equals R . However,the partition function and the integral are hard to evaluate, and approximations must in practice be introduced.In case of the weak string, Geiger and Girosi [15] used the so-called saddle point approximation while Bilbro et al. [14] used Peierl’s inequality to derive identical approximations of effective potential to be minimized. In these MF approximations, the smoothness term reads:f x T eT x s l sl s b g=-+F HG G I KJ J -221222log This term has the following properties [11]: the lower bound on the second derivative increases towards approximately -0.6l 2when s tends towards infinity. For large s the minimum in secondderivative occurs at x µ±s while for small s it occurs at x Ϸ T .T HEOREM 5 [MFA, W EAK S TRING C ONVEXITY ]. The above MeanField approximation of the weak string leads to a convex smooth-ness term for sufficiently large s if l < 0.9 or the image gradient—R Œ G is bounded from below and above .P ROOF . This follows from Theorem 1 and the above two proper-ties of f s .In practice, images are bounded (e.g., R Œ [0; I max ]) so that the finite difference approximation of the derivative is also bounded (e.g., G = [-I max ; I max ]). In this way the MFA creates a convex func-tional if s initially is chosen sufficiently large. However, the non-convexities (potential barriers) that may not be overcome travels in the solution space as —R = ±h (s ), where h (s ) decreases from I max toward T during the annealing. In this way the solution may be trapped between the potential barriers creating solutions with very low gradients. This behavior is generic for the proposed MFA. In Fig. 4 this is illustrated. Parameters (l , T ) are chosen to emphasize the trend; often the energy will drop before it increases as a function of the initial temperature.The MFA of the weak string does not gradually introduce nonconvexities, but merely intro-duces them at the boundary of the solution space and moves them inwards.Fig. 4. Phone image [15]. Left is MFA (l = 19, T = 1 like in [15]) for initial temperature s 0 = 10n, where n Œގ running from one to eight is the image number starting from upper left. All annealings end at s = 0.1. Below is the energy of the solution as a function of initial tem-perature compared to the SF and BZ-GNC solutions.6C ONCLUSIONI have derived criteria for functionals with quadratic data terms being convex. Two methods of approximating smoothness terms both based on Gaussian smoothing have been proposed. Gaussian smoothing has the advantage of being causal (i.e., not creating structure in the functional when applied [20]) and implying a semigroup structure. Applied to surface reconstruction problems, the functional approximation has produced GNC algorithms that perform marginally to significantly better than the Blake-Zisserman GNC on the weak string.The methods of approximation guarantee an initially convex functional and are fully automatic. They also provide a conserva-tive measure of the initial degree of smoothing needed. In this paper results are shown for the weak string and the robust Lorentzian estimator for reconstruction. In [18], more results can be seen. In [21], the Probability Focusing has been used in con-junction with an adaptive reconstruction scheme. Here, the prior is given as a measured histogram of gradients. Both the schemes are applicable to such situations of numerically represented smooth-ness functions.The methods have also been used in conjunction with stereo. However, in this case the data term is not quadratic and a convex functional cannot be guaranteed. Actually, it can be argued that an unbiased method can never guarantee a convex solution space in the stereo case [22].The criterion on the smoothness functional to imply a convex solution space is used to show that the Mean Field approximation of the weak string [14], [15] does not necessarily imply a convex functional. Furthermore, the MF Annealing is analyzed from the motion of nonconvexities indicating that the initial temperature defines the discontinuities and that they are arbitrarily few for a sufficiently high starting temperature.A CKNOWLEDGMENTSThe author thanks S.I. Olsen and J. Zerubia for discussions and comments and D. Geiger for making the Phone Image available.R EFERENCES[1] J. Rissanen, Stochastic Complexity in Statistical Inquiry. World Sci-entific Publishing, 1989.[2] M. Li and P. Vitányi, An Introduction to Kolmogorov Complexity andIts Applications. New York: Springer-Verlag, 1993.[3] A.N. Tikhonov and V.Y. Arsenin, Solution of Ill-Posed Problems.Washington, D.C.: Winston and Wiley, 1977.[4] M. Unser, A. Aldroubi, and M. Eden, “Recursive RegularizationFilters: Design, Properties, and Applications,” Transactions on Pat-tern Analysis and Machine Intelligence, vol. 13, Mar. 1991.[5] M. Nielsen, L. Florack, and R. Deriche, ”Regularization, ScaleSpace, and Edge Detection Filters,” J. Mathematical Imaging and Vi-sion, in press.[6] A. Rangarajan and R. Chellappa, “Generalized Graduated Non-Convexity Algorithm for Maximum A Posteriori Image Estima-tion,” Proc. 10th ICPR, Atlantic City, N.J., USA, June 1990.[7] S. Geman and D. Geman, “Stochastic Relaxation, Gibbs Distribu-tion, and the Bayesian Restoration of Images,” Transactions on Pat-tern Analysis and Machine Intelligence, vol. 6, pp. 721-741, 1984. [8] A.R.P. Meer, D. Mintz, and D.Y. Kim, “Robust Regression Meth-ods for Computer Vision: A Review,” IJCV, vol. 6, no. 1, pp. 59-70, 1991.[9] M. Nielsen, “Isotropic Regularization,” Proc. Fourth BMVC, Guild-ford, England, Sept. 21-23, 1993.[10] P. Belhumeur, “A Binocular Stereo Algorithm for ReconstructingSloping, Creased, and Broken Surfaces, in the Presence of Half-Occlusion,” Int’l Conf. Computer Vision, Berlin, 1993.[11] M. Nielsen, “Surface Reconstruction: GNCs and MFA,” Tech.Rep. 2353, Institut National de Recherche en informatique et en automatique, Centre de Diffusion, INRIA, BP 105-78153 Le Ches-nay Cedex, France, Sept. 1994.[12] R. Thom, Structural Stability and Morphogenesis, translation byD.H. Fowler. New York: Benjamin–Addison Wesley, 1975.[13] P.T. Saunders, An Introduction to Catastrophe Theory. Cambridge,England: Cambridge Univ. Press, 1980.[14] G.L. Bilbro, W.E. Snyder, S.J. Garnier, and J.W. Gault, “MeanField Annealing: A Formalism for Constructing GNC-Like Algo-rithms,” IEEE Trans. Neural Networks, vol. 3, Jan. 1992.[15] D. Geiger and F. Girosi, “Parallel and Deterministic AlgorithmsFrom MFRs: Surface Reconstruction,” Transactions on Pattern Analysis and Machine Intelligence, vol. 13, May 1991.[16] R. March, “Visual Reconstruction With Discontinuities UsingVariational Methods,” Image and Vision Computing, vol. 10, Jan.-Feb. 1992.[17] A. Blake and A. Zisserman, Visual Reconstruction. Cambridge,Mass.: MIT Press, 1987.[18] M. Nielsen, From Paradigm to Algorithms in Computer Vision, PhDthesis, Datalogisk Institut ved Kobenhavns Universitet, Copenha-gen, Denmark, Apr. 1995.[19] M. Nielsen, “Surface Reconstruction: GNCs and MFA,” Proc. Int’lConf. Computer Vision, Cambridge, Mass., USA, pp. 344-349, June 20-23, 1995.[20] J.J. Koenderink, “The Structure of Images,” Biol. Cybern., vol. 50,pp. 363-370, 1984.[21] M. Nielsen, “Adaptive Regularization: Towards Self-CalibratedSurface Reconstruction,” Tech. Rep. 2351, Institut National de Re-cherche en informatique et en automatique, Centre de Diffusion, INRIA, BP 105-78153 Le Chesnay Cedex, France, Sept. 1994. [22] M. Nielsen and R. Deriche, “Binocular Dense Depth Reconstruc-tion Using Isotropy Constraint,” G. Borgefors, ed. Theory and Ap-plications of Image Processing II—Selected Articles From the Ninth Scandinavian Conference on Image Analysis. World Scientific Pub-lishing, 1995.。

GFM包的中文标题:通用因子模型(版本1.2.1)说明书

Package‘GFM’August11,2023Type PackageTitle Generalized Factor ModelVersion1.2.1Date2023-08-10License GPL-3Author Wei Liu[aut,cre],Huazhen Lin[aut],Shurong Zheng[aut],Jin Liu[aut],Jinyu Nie[aut]Maintainer Wei Liu<********************>Description Generalized factor model is implemented for ultra-high dimensional data with mixed-type variables.Two algorithms,variational EM and alternate maximization,are designed to implement the gen-eralized factor model,respectively.The factor matrix and loading matrix together with the number of fac-tors can be well estimated.This model can be employed in social and behavioral sciences,economy andfinance,and ge-nomics,to extract interpretable nonlinear factors.More details can be referred toWei Liu,Huazhen Lin,Shurong Zheng and Jin Liu.(2021)<doi:10.1080/01621459.2021.1999818>. URL https:///feiyoung/GFMBugReports https:///feiyoung/GFM/issuesDepends doSNOW,parallel,R(>=3.5.0)Imports MASS,stats,irlba,Rcpp,methodsSuggests knitr,rmarkdownLinkingTo Rcpp,RcppArmadilloVignetteBuilder knitrEncoding UTF-8RoxygenNote7.1.11NeedsCompilation yesRepository CRANDate/Publication2023-08-1106:22:04UTCR topics documented:chooseFacNumber (2)Factorm (4)gendata (5)gfm (6)measurefun (8)overdispersedGFM (9)OverGFMchooseFacNumber (11)Index13 chooseFacNumber Choose the Number of factors for Generalized Factor ModelsDescriptionThis function is designed to chooose the number of factors for a generalized factor model.UsagechooseFacNumber(XList,types,q_set=2:10,select_method=c("SVR","IC"),offset=FALSE,dc_eps=1e-4,maxIter=30,verbose=TRUE,parallelList=NULL) ArgumentsXList a list consisting of matrices with the same rows n,and different columns(p1,p2, ...,p_d),observational mixed data matrix list,d is the types of variables,p_j isthe dimension of varibles with the j-th type.types a d-dimensional character vector,specify the type of variables.For example, types=c( gaussian , poisson , binomial ),implies the components ofXList are matrices with continuous,count and binomial values,respectively.q_set a positive integer vector,specify the candidates of factor number q,(optional) default as c(2:10)according to Bai(2013).select_method a string,specify the method to choose the number of factors.Two methods are supported:the singular value ratio(SVR)and information criterion(IC)based methods,default as’SVR’.Empirically,’SVR’is much faster than’IC’,especially for high-dimensional large-scale data.offset a logical value,whether add an offset term(the total counts for each row in the count component of XList)when there are Poisson variables.dc_eps positive real number,specify the tolerance of varing quantity of objective func-tion in the algorithm.Optional parameter with default as1e-4.maxIter a positive integer,specify the times of iteration.Optional parameter with default as50.verbose a logical value,specify whether ouput the information in iteration process,(op-tional)default as TRUE.parallelList a list with two components:(1)parallel:a logical value with TRUE or FALSE,indicates wheter to use prallelcomputating.Optional parameter with default as FALSE.(2)ncores:a positive integer,specify the number of cores when parallel comput-ing is used.This argument plays its role if only select_method= IC .Valuereturn an integer value,the estimated number of factors.NotenothingAuthor(s)Liu WeiReferencesLiu,W.,Lin,H.,Zheng,S.,&Liu,J.(2021).Generalized factor model for ultra-high dimen-sional correlated variables with mixed types.Journal of the American Statistical Association,(just-accepted),1-42.See AlsonothingExamples##mix of normal and Poissondat<-gendata(seed=1,n=60,p=60,type= norm_pois ,q=2,rho=2)##we set maxIter=2for example.hq<-chooseFacNumber(dat$XList,dat$types,verbose=FALSE,maxIter=2)4Factorm Factorm Factor Analysis ModelDescriptionFactor analysis to extract latent linear factor and estimate loadings.UsageFactorm(X,q=NULL)ArgumentsX a n-by-p matrix,the observed dataq an integer between1and p or NULL,default as NULL and automatically choose q by the eigenvalue ratio method.Valuereturn a list with class named fac,including following components:hH a n-by-q matrix,the extracted lantent factor matrix.hB a p-by-q matrix,the estimated loading matrix.q an integer between1and p,the number of factor extracted.sigma2vec a p-dimensional vector,the estimated variance for each error term in model.propvar a positive number between0and1,the explained propotion of cummulative variance by the q factors.egvalues a n-dimensional(n<=p)or p-dimensional(p<n)vector,the eigenvalues of sample covariance matrix.NotenothingAuthor(s)Liu WeiReferencesFan,J.,Xue,L.,and Yao,J.(2017).Sufficient forecasting using factor models.Journal of Econo-metrics.See Alsogfm.gendata5 Examplesdat<-gendata(n=300,p=500)res<-Factorm(dat$X)measurefun(res$hH,dat$H0)#the smallest canonical correlationgendata Generate simulated dataDescriptionGenerate simulated data from high dimensional genelized nonlinear factor model.Usagegendata(seed=1,n=300,p=50,type=c( homonorm , heternorm ,pois , bino , norm_pois , pois_bino , npb ),q=6,rho=1,n_bin=1)Argumentsseed a nonnegative integer,the random seed,default as1.n a positive integer,the sample size.p an positive integer,the variable dimension.type a character,specify the variables types for generated data,default as’homonorm’, representing the homogeneous gaussian variables.q a positive integer,the number of factors.rho a positive number,controlling the magnitude of loading matrix.n_bin a positive integer,specify the number of trails for the binomial variables when type is set to one of’bino’,’pois_bino’and’npb’.DetailsThis function provides a variaty of mix of different variable types,in which’homonorm’repre-sents the generated data with only homogenous normal variables;’heternorm’represents the data with only heterogenous normal variables;’pois’means the data with only poisson variables;’bino’means the data with only binomial variables;’norm_pois’means the mix of normal and poisson variables;’pois_bino’represents the mix of poisson and binomial variables;and’npb’means the most complex mix of normal,poisson and binomial variables.6gfm Valuereturn a list including two components:X a n-by-p matrix,the observed data matrix.XList a list consisting of the above observed data matrices with the same rows n(ob-servations),and different columns(p1,p2,...,p_d)and p columns in total,whered is the types of variables,pj is the dimension of varibles with the j-th type.H0a n-by-q matrix,the true lantent factor matrix.B0a p-by-q matrix,the true loading matrix,the last pzero rows are vectors of zeros.mu0a p-dimensional vector,the true intercept terms.NotenothingAuthor(s)Wei LiuSee AlsoFactorm;gfm.Examplesdat<-gendata(n=300,p=500)str(dat)gfm Generalized Factor ModelDescriptionThis function is to implement the generalized factor model.Usagegfm(XList,types,q=10,offset=FALSE,dc_eps=1e-4,maxIter=30,verbose=TRUE,algorithm=c("VEM","AM"))gfm7ArgumentsXList a list consisting of matrices with the same rows n,and different columns(p1,p2, ...,p_d),observational mixed data matrix list,d is the types of variables,p_j isthe dimension of varibles with the j-th type.types a d-dimensional character vector,specify the type of variables.For example, types=c( gaussian , poisson , binomial ),implies the components ofXList are matrices with continuous,count and binomial values,respectively.q a positive integer or empty,specify the number of factors,defualt as10.offset a logical value,whether add an offset term(the total counts for each row in the count component of XList)when there are Poisson variables.dc_eps a positive real,specify the relative tolerance of objective function in the algo-rithm.Optional parameter with default as1e-4.maxIter a positive integer,specify the times of iteration.Optional parameter with default as30.verbose a logical value with TRUE or FALSE,specify whether ouput the information in iteration process,(optional)default as TRUE.algorithm a string,specify the algorithm to be used forfitting model.Now it supports two algorithms:variational EM(VEM)and alternate maximization(AM)algorithm,default as VEM.Empirically,we observed that VEM is more robust than AM tothe high noise data.DetailsThis function also has the MATLAB version at https:///feiyoung/MGFM/blob/ master/gfm.m.Valuereturn a list with class name’gfm’and including following components,hH a n*q matrix,the estimated factor matrix.hB a p*q matrix,the estimated loading matrix.hmu a p-dimensional vector,the estimated intercept terms.obj a real number,the value of objective function when the convergence achieves.q an integer,the used or estimated factor number.history a list including the following7components:(1)dB:the varied quantity of B in each iteration;(2)dH:the varied quantity of H in each iteration;(3)dc:thevaried quantity of the objective function in each iteration;(4)c:the objectivevalue in each iteration;(5)realIter:the real iterations to converge;(6)maxIter:the tolerance of maximum iterations;(7)elapsedTime:the elapsed time.Notenothing8measurefunAuthor(s)Liu WeiReferencesLiu,W.,Lin,H.,Zheng,S.,&Liu,J.(2021).Generalized factor model for ultra-high dimen-sional correlated variables with mixed types.Journal of the American Statistical Association,(just-accepted),1-42.Bai,J.and Liao,Y.(2013).Statistical inferences using large esti-mated covariances for panel dataand factor models.See AlsonothingExamples##mix of normal and Poissondat<-gendata(seed=1,n=60,p=60,type= norm_pois ,q=2,rho=2)##we set maxIter=2for example.gfm2<-gfm(dat$XList,dat$types,q=2,verbose=FALSE,maxIter=2)measurefun(gfm2$hH,dat$H0,type= ccor )measurefun(gfm2$hB,dat$B0,type= ccor )measurefun Assess the performance of an estimator on a matrixDescriptionEvaluate the smallest cononical correlation(ccor)coefficients or trace statistic between two matri-ces,where a larger ccor or trace statistic is better.Usagemeasurefun(hH,H,type=c( trace_statistic , ccor ))ArgumentshH a n-by-q matrix,the estimated matrix.H a n-by-q matrix,the true matrix.type a character taking value within c( trace_statistic , ccor ),default as’trace_statistic’. Valuereturn a real number.NotenothingAuthor(s)Liu WeiExamplesdat<-gendata(n=100,p=200,q=2,rho=3)res<-Factorm(dat$XList[[1]])measurefun(res$hB,dat$B0)overdispersedGFM Overdispersed Generalized Factor ModelDescriptionThis function is to implement the overdispersed generalized factor model.UsageoverdispersedGFM(XList,types,q,offset=FALSE,epsELBO=1e-5,maxIter=30,verbose=TRUE)ArgumentsXList a list consisting of matrices with the same rows n,and different columns(p1,p2, ...,p_d),observational mixed data matrix list,d is the types of variables,p_j isthe dimension of varibles with the j-th type.types a d-dimensional character vector,specify the type of variables.For example, types=c( gaussian , poisson , binomial ),implies the components ofXList are matrices with continuous,count and binomial values,respectively.q a positive integer or empty,specify the number of factors.offset a logical value,whether add an offset term(the total counts for each row in the count component of XList)when there are Poisson variables.epsELBO a positive real,specify the relative tolerance of ELBO function in the algorithm.Optional parameter with default as1e-5.maxIter a positive integer,specify the times of iteration.Optional parameter with default as30.verbose a logical value with TRUE or FALSE,specify whether ouput the information in iteration process,(optional)default as TRUE.DetailsOverdispersion is prevalent in practical applications,particularly infields like biomedical and ge-nomics studies.To address this practical demand,we propose an overdispersed generalized fac-tor model(OverGFM)for performing high-dimensional nonlinear factor analysis on overdispersed mixed-type data.Valuereturn a list with class name’overdispersedGFM’and including following components,hH a n*q matrix,the estimated factor matrix.hB a p*q matrix,the estimated loading matrix.hmu a p-dimensional vector,the estimated intercept terms.obj a real number,the value of objective function when the convergence achieves.q an integer,the used or estimated factor number.history a list including the following7components:(1)dB:the varied quantity of B in each iteration;(2)dH:the varied quantity of H in each iteration;(3)dc:thevaried quantity of the objective function in each iteration;(4)c:the objectivevalue in each iteration;(5)realIter:the real iterations to converge;(6)maxIter:the tolerance of maximum iterations;(7)elapsedTime:the elapsed time.NotenothingAuthor(s)Liu WeiSee AlsonothingExamples##mix of normal and Poissondat<-gendata(seed=1,n=60,p=60,type= norm_pois ,q=2,rho=2)##we set maxIter=2for example.gfm2<-overdispersedGFM(dat$XList,dat$types,q=2,verbose=FALSE,maxIter=2)measurefun(gfm2$hH,dat$H0,type= ccor )measurefun(gfm2$hB,dat$B0,type= ccor )OverGFMchooseFacNumberChoose the Number of factors for Overdispersed Generalized FactorModelsDescriptionThis function is designed to chooose the number of factors for the overdispersed generalized factormodel by using the singular value ratio(SVR)based method.UsageOverGFMchooseFacNumber(XList,types,q_max=15,offset=FALSE,epsELBO=1e-4,maxIter=30,verbose=TRUE,threshold=1e-2)ArgumentsXList a list consisting of matrices with the same rows n,and different columns(p1,p2,...,p_d),observational mixed data matrix list,d is the types of variables,p_j isthe dimension of varibles with the j-th type.types a d-dimensional character vector,specify the type of variables.For example,types=c( gaussian , poisson , binomial ),implies the components ofXList are matrices with continuous,count and binomial values,respectively.q_max a positive integer,specify the upper bound of the number of factors,defualt as15.offset a logical value,whether add an offset term(the total counts for each row in thecount component of XList)when there are Poisson variables.epsELBO a positive real,specify the relative tolerance of ELBO function in the algorithm.Optional parameter with default as1e-5.maxIter a positive integer,specify the times of iteration.Optional parameter with defaultas30.verbose a logical value with TRUE or FALSE,specify whether ouput the information initeration process,(optional)default as TRUE.threshold a postive real,the threshold that is used tofilter the small singular values in theSVR criterion.Valuereturn an integer value,the estimated number of factors.NotenothingAuthor(s)Liu WeiSee AlsonothingExamples##mix of normal and Poissondat<-gendata(seed=1,n=60,p=60,type= norm_pois ,q=2,rho=2)##we set maxIter=2for example.hq<-OverGFMchooseFacNumber(dat$XList,dat$types,verbose=FALSE,maxIter=2)IndexchooseFacNumber,2Factorm,4,6gendata,5gfm,4,6,6measurefun,8overdispersedGFM,9 OverGFMchooseFacNumber,1113。

  1. 1、下载文档前请自行甄别文档内容的完整性,平台不提供额外的编辑、内容补充、找答案等附加服务。
  2. 2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
  3. 3、如文档侵犯您的权益,请联系客服反馈,我们会尽快为您处理(人工客服工作时间:9:00-18:30)。

1. INTRODUCTION
The post-Gaussian trial function provides a robust instrument for obtaining the variational solutions1,2 . By using the post-Gaussian trial function, one gets an excellent estimate for the ground-state of the system. These results indicate the possibility for obtaining better solutions for the potential problems, not only the ground state but also the excited states. In this article, we examine the utility of the post-Gaussian trial wave functions. For this purpose, we employ the anharmonic oscillator as an example. The anharmonic oscillator has been given the useful laboratory to test a nonperturbative approximation. We also show explicitly that the post-Gaussian variational method gives better solutions to the anharmonic oscillator than the ordinary Gaussian one. This may give insight into a quantum-fieldtheoretical problem in the standard Gaussian variational approach to spontaneous symmetry breakings. In the following section, we solve the ground-state energy of the anharmonic oscillator 1
(0) 1,2
: (2)
=
nα 2n . Γ( 21 ) n
1
(3)
Γ(z ) is Euler’s gamma function of argument z . We notice that this trial function in Eq.(2) has two variational parameters; i.e. α and n. In the case of n = 1, this trial function reduces to the familiar Gaussian trial function which is often taught in an elementary course of quantum mechanics. Using Eq.(2), we calculate the expectation value of the Hamiltonain as follows: I (0) (α, n) ≡ ψ (0) (n, α)|H |ψ (0)(n, α) = Γ( 25 ) 1/n 1 Γ( 23 ) −1/n ) −2/n n2 Γ(2 − 21 n n n α + α + g . 1 1 1 α 2 Γ( 2n ) 2 Γ( 2n ) Γ( 2n ) (4) (5)
Post-Gaussian variational method for quantum anharmonic oscillator
arXiv:physics/9905056v1 [physics.ed-ph] 28 May 1999
Akihiro Ogura
Laboratory of Physics College of Science and Technology, Nihon University Funabashi, Chiba 274-8501, Japan
3. THE EXCITED STATES AND THE NUMERICAL RESULTS
The first excited state ψ (1) can be determind in such a way that the following orthogonality condition is satisfied: < ψ (0) (n0 , α0 )|ψ (1) (n0 , β ) >= 0, (8)
where β0 is the fixed value which make Eq.(9) optimal and γ is again a new variational parameter. The same procedure stated above can repeatedly be applied to find the higher excited states. Now, inspired by the harmonic-oscillator wave functions, we propose to take the following trial functions for the excited states:
where g is a positive constant. To obtain the ground-state energy of H using the variational method, we employ a post-Gaussian trial function α ψ (0) = N (0) exp − |x|2n , 2 where the normalization constant N (0) is given by N
2. POST-GAUSSIAN VARIATIONAL METHOD AND THE GROUND STATE
The system we consider is the anharmonic oscillator. The Hamiltonian is written as follows: H = p2 x2 + + gx4 , 2 2 (1)
4. CONCLUSIONS
We have applied the post-Gaussian trial functions to the anharmonic oscillators. We have explicitly calculated up to the second excited state and compared the results with numerical calculations. We have seen how the post-Gaussian trial functions give the better results than the Gaussian wavefunctions. Finally, we note that Cooper et al.4,5 have also discussed the variational energy eigenvalues for the anharmonic oscillator using the post-Gaussian trial functions. Their algorithm to obtain the excited states is based on the idea of SUSY quantum mechanics6 , whereas the method we discussed here is the standard one known in elementary quantum mechanics.
The optimization of Eq.(5) with respect to the two parameters α and n ∂I (0) ∂α ∂I (0) ∂n = 0,
α0
(6) (7) 2
= 0,
n0
can easily accomplished numerically. In Table 1., we show the numerical result for the ground state. The optimal value is found to be at n = n0 = 1.13493 that indicates the wave function deviates from the Gaussian one.
using the variational method with post-Gaussian trial wave function. In section 3, we calculate some excited-state energy levels and compare the results with numerical calculations. Section 4 is devoted to discussions. Throughout this paper all quantities and variables are assumed to be made dimensionless for simplicity.
Abstract
Using the post-Gaussian trial functions, we calculate the variational solutions to the quantum-mechanical anharmonic oscillator. We evaluate not only the ground state but also some excited energies, and compare them with numerical results.
相关文档
最新文档