10计算材料物理-第四章

计算材料物理专题结构搜索和预测2WhyisStructurePredictionHard?EnergyAtomicpositionsLocalminimaGlobalminimumTruestructureAccuratePotentialEnergySurface----abinitioHugenumberoflocalminima----how?随机抽样方法RandomSamplingmethodsinvolveintherandomgenerationofalargenumberofstructuresonPES;AIRSS(abinitiorandomstructuresearching)~ajm/airss/airss.html翻越势垒方法模拟退火算法(simulatedannealing;SA)盆地跳算法(Basin-hopping)minimahoppingmetadynamicsalgorithm模拟退火算法Basinhoppingminimahoppingmetadynamics演化(进化)算法遗传算法(geneticalgorithms;GA)粒子群优化算法(particleswarmoptimization;PSO)蚁群优化算法(antcolonyoptimization;ACO)遗传算法SwarmIntelligence群体智能Swarm可被描述为一些相互作用相邻个体的集合体，蜂群、蚁群、鸟群都是Swarm的典型例子。鱼聚集成群可以有效地逃避捕食者，因为任何一只鱼发现异常都可带动整个鱼群逃避。蚂蚁成群则有利于寻找食物，因为任一只蚂蚁发现食物都可带领蚁群来共同搬运和进食。一只蜜蜂或蚂蚁的行为能力非常有限，它几乎不可能独立存在于自然世界中，而多个蜜蜂或蚂蚁形成的Swarm则具有非常强的生存能力，且这种能力不是通过多个个体之间能力简单叠加所获得的。社会性动物群体所拥有的这种特性能帮助个体很好地适应环境，个体所能获得的信息远比它通过自身感觉器官所取得的多，其根本原因在于个体之间存在着信息交互能力。粒子群优化算法（ParticleSwarmOptimization）1995年由J.Kennedy,R.C.Eberhart等人提出该算法最初是受到鸟群活动的规律性启发，进而利用群体智能建立的一个简化模型。粒子群优化算法利用群体中的个体对信息的共享使整个群体的运动在问题求解空间中产生从无序到有序的演化过程，从而获得最优解。PSO同遗传算法类似，是一种基于迭代的优化算法。系统初始化为一组随机解，通过迭代搜寻最优值。但是它没有遗传算法用的交叉(crossover)以及变异(mutation)，而是粒子在解空间追随最优的粒子进行搜索。同遗传算法比较，PSO的优势在于简单容易实现并且没有许多参数需要调整。目前已广泛应用于函数优化，神经网络训练，模糊系统控制以及其他遗传算法的应用领域。结构搜索和预测程序AIRSSAbinitioRandomStructureSearchingGASPGeneticAlgorithmforStructureandPhasePredictionCALYPSOCrystalstructureAnaLYsisbyParticleSwarmOptimizationUSPEXUniversalStructurePredictor:EvolutionaryXtallography~ajm/airss/airss.htmlChrisJPickard://avogadro.cc/wiki/Main_PageDavidC.Lonie,EvaZurek;XtalOpt:AnOpen-SourceEvolutionaryAlgorithmforCrystalStructurePrediction,ComputerPhysicsCommunications182(2011)pp.372-387XtalOptisafreeandtrulyopensourceevolutionaryalgorithmdesignedtopredictcrystalstructures.ItisimplementedasanextensiontotheAvogadromoleculareditor.XtalOptrunsonaworkstationandsupportsusingGULP,VASP,pwSCF(QuantumESPRESSO),andCASTEPforgeometryoptimizations.StateUniversityofNewYorkatBuffaloTheGeneticalgorithmforstructureprediction–GASP–predictsthestructureandcompositionofstableandmetastablephasesofcrystals,molecules,atomicclustersanddefectsfromfirst-principles.TheGASPprogramisinterfacedtomanyenergycodesincluding:VASP,LAMMPS,MOPAC,Gulp,JDFTxandcanefficientlyrunonparallelarchitectures.CALYPSO(CrystalstructureAnaLYsisbyParticleSwarmOptimization)isanefficientstructurepredictionmethodanditssame-namecomputersoftware.TheCALYPSOpackageisprotectedbytheCopyrightProtectionCenterofChinawiththeregistrationNo.2010SR028200andclassificationNo.61000-7500.FreelydistributedonAcademicuseundertheregulationstermedintheCALYPSO_LICENCE.朱黎吕健王彦超马琰铭教授吉林大学超硬材料国家重点实验室CALYPSOWHATISTHEFEATURE?Predictionsoftheenergeticallystable/metastablestructuresatgivenchemicalcompositionsandexternalconditions(e.g.,pressure)forclusters,2Dlayers,surfaces,and3Dcrystals.Designofnovelfunctionalmaterials,e.g.,superhardmaterials.OptionsforthestructuralevolutionsusingglobalorlocalPSO.Structuresearcheswithautomaticvariationofchemicalcompositions.Structurepredictionswithfixedcellparameters,orfixedspacegroups,orfixedmolecules.CALYPSOiscurrentlyinterfacedwithGAUSSIAN,DFTB+,VASP,CASTEP,QuantumEspresso,GULP,SIESTAandCP2Kcodes.Theinterfacewithothertotalenergycodescanalsobeimplementedbyusers'request.HistoryofCALYPSOTheCALYPSOteamindependentlyinitializedtheideaofapplyingPSOalgorithmintostructurepredictionin2006(MaandWang)firstapplicationofPSOalgorithmintostructurepredictionof3DcrystalsbyWang,Lv,Zhu&Main2010,2DlayersbyWang,Miao,etal,in20122DsurfacereconstructionbyLuetal,in2014StructuresearchingefficienciesofisolatedsystemshavebeensubstantiallyimprovedbytheCALYPSOteam(Lv,Wang,Zhu&Ma)in2012,wherethesuccessofthisapplicationhasbeenbackedupwiththeintroductionofvariousefficienttechniques(e.g.,bondcharacterizationmatrixforfingerprintingstructures,symmetryconstraintsonstructuregeneration,etc.).MajorTechniquesEmployedStructuralevolutionthroughPSOalgorithm.PSOisbest-knownforitsabilitytoconquerlargebarriersofenergylandscapesbymakinguseoftheswarmintelligenceandbyself-improvingstructures.BothglobalandlocalPSOalgorithmshavebeenimplemented.TheglobalPSOhastheadvantageoffastconvergence,whilelocalPSOisgoodatavoidingprematureconvergenceandthusenhancethecapabilityofCALYPSOindealingwithmorecomplexsystems.MajorTechniquesEmployedSymmetryconstraintsduringstructuregenerationtoreducesearchingspaceandenhancethestructuraldiversity.Structuralcharacterizationtechniquestoeliminatesimilarstructures,definenonflyingareas,enhancesearchingefficiency,anddivideenergysurfacesforlocalPSOsearching.(i)bondcharacterizationmatrixtechnique(ii)atom-centeredsymmetricalfunctiontechniqueMajorTechniquesEmployedIntroducingnewstructurespergenerationwithcontrollablepercentagetoenhancestructuraldiversity.Interfacetoanumberoflocalstructuraloptimizationc