An ionically based mapping model with memory for cardiac restitution

一文详解flow based modelsFlow based models，也被称为可逆生成模型（invertible generative models），是一类用于生成模型的神经网络架构。

与其他生成模型如GAN和VAE不同，flow based models拥有可逆的编码器和解码器结构，使得输入样本可以通过解码器生成样本，同时编码器可以恢复原始样本。

本文将详细解释flow based models的原理和相关参考内容。

首先，flow based models的核心思想是建立输入和输出之间的一对一映射关系，以及通过联合分布近似来进行建模。

flow based models的主要优点是可以计算出精确的似然函数，而不需要通过变分推断或逼近技术。

同时，由于其可逆性质，flow based models还可以进行完全可解推理和采样。

具体地，flow based models通常由多个可逆层组成，每个层都有一个从输入到输出的可逆函数。

这些函数可以是简单的元素级函数如仿射变换和逐通道的非线性函数，也可以是复杂的非线性转换函数如卷积神经网络。

整个模型的可逆性由这些可逆层的组合实现。

通过这些可逆层，flow based models可以将原始样本空间映射到一个更简单的潜在空间，然后通过解码器进行重构。

近年来，flow based models在图像生成、图像修复、语言建模和强化学习等领域取得了广泛应用。

下面列举一些与flow based models相关的参考内容，供读者深入了解和学习：1. "Flow++: Improving Flow-Based Generative Models withVariational Dequantization and Architecture Design"，由Jonathan Ho等人于2019年提出的论文，介绍了一种改进的flow based model，通过使用变分量化和架构设计提高了模型的生成效果。

Electrochimica Acta52(2006)681–687Aflow model of porous anodicfilm growth on aluminiumS.J.Garcia-Vergara a,P.Skeldon a,∗,G.E.Thompson a,H.Habazaki ba Corrosion and Protection Centre,School of Materials,The University of Manchester,P.O.Box88,Manchester M601QD,UKb Graduate Engineering School,Hokkaido University,N13W8,Kita-ku,Sapporo060-8628,JapanReceived5April2006;received in revised form12May2006;accepted28May2006Available online7July2006AbstractThe development of pores in a classical porous anodicfilm formed on aluminium in phosphoric acid solution is investigated.The study employs a tungsten tracer layer that is incorporated into the anodicfilm from the aluminium substrate,followed by detection of the tracer by transmission electron microscopy and Rutherford backscattering spectroscopy.Distortions of the tungsten layer on entry into thefilm and retention of tungsten species in thefilm are compatible with porosity arising mainly fromflow of anodic oxide beneath the pore bases towards the cell walls.The behaviour is contrary to expectations of a dissolution model of pore formation.©2006Elsevier Ltd.All rights reserved.Keywords:Aluminium;Anodizing;Anodic oxide;Porousfilm1.IntroductionThe interest in the present study lies in the formation of porous morphologies in anodicfilms,which are of major importance in the surface treatment of aluminium[1].Thefilms are composed mainly of amorphous alumina,often with a specific influence of the electrolyte,such as the incorporation of sulphur and phos-phorus species intofilms formed in sulphuric and phosphoric acid solutions,respectively[2,3].Thefilms comprise a rela-tively thin barrier layer next to the metal and an outer porous layer[4,5],with approximately cylindrical pores extending from thefilm surface to the barrier layer.Each pore is located within a cell of anodic alumina,with groups of cells having a potential for organization into regular patterns[6–9].The thickness of the barrier layer,the diameter of the pores and the size of the cells correlate with the anodizing voltage,with typical ratios of the order1nm V−1,the precise values depending upon parameters such as current density and composition of the electrolyte.On the other hand,the porous layer thickens in proportion to the charge passed during anodizing at a particular current density. The barrier layer at the base of each cell is approximately hemi-spherical,such that the metal/oxide interface is scalloped.The voltage during steady porousfilm growth,and hence the size ∗Corresponding author.Tel.:+441613064872;fax:+441613064865.E-mail address:p.skeldon@(P.Skeldon).offilm features,depends upon the anodizing conditions,partic-ularly the nature of the electrolyte.Pore generation is usually attributed to a thermally-assisted,field-accelerated dissolution of anodic alumina at the base of each pore[10].The growth of the oxide occurs mainly at the metal/oxide interface due to inward migration of O2−ions across the barrier layer[11,12]. Simultaneously,Al3+ions migrate outward and are ejected to the electrolyte at the pore base.The O2−and Al3+ions contribute about60and40%of the ionic current in the barrier layer for anodizing at constant voltage in sulphuric acid solution at287K.In the present work,the development of pores in anodic alu-mina is examined by introduction into thefilm of afine band of tungsten tracer,which is then followed as it moves through thefilm.Thefilm was formed in phosphoric acid solution,an electrolyte employed in production of self-ordered porosity[13]. The distortion of the tracer distribution and the absence of major loss of tracer to the electrolyte indicate that the pores are created mainly byflow of material from the pore bases to the cell walls. The material displacement is facilitated by the plasticity of the film material duringfilm growth,which arises from the ionic transport processes in thefilm.2.ExperimentalAluminium was deposited by magnetron sputtering,to about 580nm thickness,upon electropolished aluminium substrates,0013-4686/$–see front matter©2006Elsevier Ltd.All rights reserved. doi:10.1016/j.electacta.2006.05.054682S.J.Garcia-Vergara et al./Electrochimica Acta52(2006)681–687of dimensions3cm×1cm,that had been anodized to150V at 5mA cm−2in0.1M ammonium pentaborate solution at293K. The resultant,180nm-thick,barrierfilms provide position refer-ences in specimens[14].The sputtering was carried in an Atom Tech system,with targets of99.99%aluminium and99.9%tung-sten,each of50mm diameter.The system wasfirst evacuated to3×10−7mbar,with subsequent deposition of aluminium at 320mA and345V in99.999%argon at5×10−3mbar.The tungsten target was activated briefly to form a tracer layer at the middle of the deposited aluminium.The substrates were attached to a copper table,with the temperature remaining below 305K.Porous anodicfilms were then formed by anodizing at 5mA cm−2in0.4M phosphoric acid solution at293K.A two electrode cell was employed,with an aluminium cathode.Prior to anodizing,each substrate was cut to provide two approxi-mately equal pieces.One piece was retained,while the other piece was anodized.V oltage–time responses were recorded dur-ing anodizing.Ultramicrotomed sections,with nominal thickness of15nm, of the two pieces of each specimen were examined by trans-mission electron microscopy(TEM),using a JEOL2000FX II instrument operated at100kV.Tungsten contents were quanti-fied by Rutherford backscattering spectroscopy(RBS),employ-ing2.0MeV He+ions produced by the Van de Graaff accelerator of the University of Paris,with detection of scattered particles at 165◦to the direction of the incident beam.The data were inter-preted using the RUMP program.The efficiency of anodizing was determined from the oxygen contents of specimens,mea-sured to an accuracy of3%by nuclear reaction analysis(NRA), using880keV2H+ions,with detection of emitted protons at 150◦to the direction of the incident beam[15].Comparisons of signals from anodicfilms were made with that from an anodized tantalum standard of known oxygen content.3.Results3.1.Voltage–time behaviourThe voltage–time response for the aluminium layer,which was highly reproducible,was typical of the anodizing condi-tions,with an initial voltage rise at about1.0V s−1,followed by transition to a relatively steady voltage in the range100–112V (Fig.1),associated with establishment of the major pores.The steep slope at about350s indicates complete anodizing of the layer.A broad peak,between about160and265s,with height of about10V,identifies oxidation of the tracer,as evident in later TEM.The peak extends from about46to76%,of the total time of anodizing,with the peak maximum at59%.3.2.Morphologies of the substrate and anodicfilms from transmission electron microscopyA transmission electron micrograph reveals the sputtering-deposited aluminium of580±10nm thickness,with columnar grains of width about50–100nm(Fig.2(a)).The columnar grain structure is typical of the conditions of sputtering-deposition. The3–5nm thick tracer layer,evident due to atomicnumber Fig.1.V oltage–time response for anodizing the sputtering-deposited aluminium layer,with an incorporated tungsten tracer layer,at5mA cm−2in0.4M phos-phoric acid solution at293K.contrast,is located at an average depth of49%of the aluminium thickness,with variation between depths of about45–55%of the layer thickness,due to the roughness of the aluminium during sputtering.The lower limit of the range agrees with the start of the peak in the voltage–time response,i.e.at46%of the total anodizing time.The upper limit is below the peakfinish at76%of the total anodizing time,but is reasonably close to the peak maximum,of59%,and evidently indicates oxidation of most of the tungsten.The subsequent tail on the peak is due to influences of the tracer located within thefilm.The charge passed during anodizing,1.75C cm−2,agrees to about4%with the charge required for anodizing the580nm aluminium layer, about1.68C cm−2.A transmission electron micrograph of an anodicfilm formed for180s,corresponding to the start of tracer oxidation,dis-closed the tungsten band,up to20nm thick,just entering the film beneath pore bases,while still remaining in the substrate near the cell boundary regions(Fig.2(b)).The tracer in thefilm is about30–60nm distant from the metal,with the maximum distance below the pores.The apparent band dimensions are probably influenced by the distribution of tungsten through the section thickness.The curvature of the tracer band is slightly less than that of the metal/film interface,apart from the edges of the band,which bend sharply toward the metal.Fine,incipient pores near thefilm surface are formed before the establishment of the major pores.The barrier layer thicknesses for the present films correspond to a formation ratio of about1.1nm V−1,from measurements on sections that intersect the centres of pores.The growth rates of thefilms were about2.3nm s−1.Previous work, using conditions of anodizing similar to those of the present work,indicates pore diameters corresponding to a formation ratio of about1.29nm V−1[5].Following anodizing for240s,just past the peak maximum, the mean thickness of thefilm was about540nm(Fig.2(c)).The main tracer band,of width up to about20nm,is now located fully within the barrier layer,typically about the mid-thickness, suggesting a further displacement of up to about30nm from the metal/oxide interface.In traversing a cell,the distance of theS.J.Garcia-Vergara et al./Electrochimica Acta52(2006)681–687683Fig.2.Transmission electron micrographs of(a)the sputtering-deposited aluminium,with an incorporated tungsten tracer layer,and following anodizing for(b) 180,(c)240and(d)350s at5mA cm−2in0.4M phosphoric acid solution at293K.tracer from the metal increases near the cell boundary,reaches a peak within the cell wall,then falls sharply toward the middle of the barrier layer beneath the pore,where the band is much fainter,due to a reduced concentration of tungsten.The tung-sten at the pore base is about70–80nm below the tungsten in the cell wall.Notably,the depth variations of the tracer in the deposited aluminium cannot account for the consistent distribu-tions of tungsten in all cells.Further,the local depth variations within the alloy,of about30nm with respect to the mean depth, would lead to typical depth variations of up to50nm in the cells,assuming a Pilling-Bedworth ratio of Al/Al2O3of1.65, corresponding to afilm density of about3.1g cm−3[16].Fur-ther,such variations would also sometimes oppose the observed trends.By350s,thefilm is770–800nm thick(Fig.2(d)).Only occa-sional residues of aluminium remained in interstices at thefilm base.The tracer is located in a main band at a depth of38–50%of thefilm,with a mean depth of about45%compared with49%in the initial aluminium.Migration of tungsten in the porous film is only significant in the high-field,barrier region.Tung-sten species usually migrate at about30%of the rate of Al3+ ions in barrier anodicfilms[17,18];thus,tungsten may be antic-ipated to move by roughly35nm,which is close to the measured displacement of about31nm.The former movement is deter-mined from the barrier layer thickness of about115nm,with outward migration of tungsten species for30%of this distance. The measured value corresponds to4%of the anodicfilm thick-ness,representing the displacement of the tungsten from a depth of49%of the aluminium to45%of thefilm.In addition to the main tracer band,extensions along cell boundaries are often evi-dent for distances of about100–200nm.These are more distinct in a backscattered scanning electron micrograph of the block from which the TEM sections were cut(Fig.3).The thicknesses of this and the previousfilms exceed that of the oxidized alu-minium by a factor in the range1.32–1.38,similar to previous findings[14].684S.J.Garcia-Vergara et al./Electrochimica Acta52(2006)681–687Fig. 3.Scanning electron micrograph(backscattered electrons)of the sputtering-deposited aluminium layer,with an incorporated tungsten tracer layer, following anodizing for350s at5mA cm−2in0.4M phosphoric acid solution at293K.3.3.Film compositions from Rutherford backscattering spectroscopyRBS spectra disclosed6.6×1015W atoms cm−2in the ini-tial aluminium,equivalent to an average composition of aboutAl–30at.%W,using the average thickness of the tracer layerfrom TEM and the weighted atomic densities of aluminium andtungsten(Fig.4).The tungsten peaks are broadened followinganodizing due mainly to a combination of the altered tungstendistribution and the influences of porosity on the paths of ionsthrough thefilm,which will affect the amount offilm materialtraversed by the ions in penetrating to a given depth of thefilm.The composition of the mainfilm material can be expressed asAl2O3·0.116AlPO4,as found previously[14].The porousfilm material overlying the tungsten tracer,which is formed in theearlier stages of anodizing,is essentially the same for all speci-mens,such that its influences on measured contents of tungstenare expected to be similar.Notably,the integrated peak yieldsof tungsten were similar,with no systematic trend,for speci-mens anodized180,200,220,240,260and350s,representingtimes from the initial anodizing of the tracer to the completeanodizing of the aluminium,with a maximum deviation of anyvalue from the mean of7%.Further,comparison with analysesof the initial substrates used for anodizing times of240and350srevealed ratios of tungsten yields prior to and following anodiz-ing of0.98and1.00,respectively,to an accuracy of about4%,primarily associated with uncertainty of the background signal.A reduction by a factor of0.96is expected from the peak shiftto lower energies following anodizing,and hence the increasedscattering cross-section.Evidently,thefindings are consistentwith negligible or minor losses of tungsten as a consequence ofanodizing.3.4.Efficiency of anodizing from nuclear reaction analysisThe oxygen contents of specimens anodized in phosphoricacid solution increase linearly with the anodizing time,at arate Fig.4.(a)RBS spectra for the sputtering-deposited aluminium layer,with an incorporated tungsten tracer layer,before and after anodizing for350s at 5mA cm−2in0.4M phosphoric acid solution at293K.(b)Details of the tung-sten peaks,including the peak for an anodizing time of180s.of1.06×1016oxygen atoms cm−2s−1(Fig.5).This rate cor-responds to an anodic current density of3.1mA cm−2,using an average charge number for the oxygen ions of−1.83that accounts for the incorporation of phosphate ions into thefilm. An efficiency offilm growth of62%is indicated from compar-ison with the actual anodizing current density of5mA cm−2.4.Discussion4.1.Growth mechanism of porous anodicfilmsThe present porous anodicfilms develop their characteristic morphologies before the tracerfirst enters thefilm,as units of WO3[17,19]within the anodic alumina.Once the metal/oxide interface reaches the tracer band,tungsten is oxidized withS.J.Garcia-Vergara et al./Electrochimica Acta52(2006)681–687685Fig.5.Dependence of oxygen contents of anodicfilms on anodizing time for the sputtering-deposited aluminium layer,with an incorporated tungsten tracer layer,following anodizing at5mA cm−2in0.4M phosphoric acid solution at 293K.negligible requirement for prior enrichment of tungsten in the aluminium[17,20].Such enrichment is a necessary pre-cursor to oxidation of tungsten in relatively dilute Al–W alloys,with required levels of enrichment corresponding to an average com-position in the enriched alloy layer close to that of the present tracer layer.Following the conventional model of porousfilm growth,tungsten and aluminium species migrate toward the film/electrolyte interface,wherefield-assisted ejection and dis-solution processes take place that maintain a constant thickness of the barrier layer.Thefirst andfinal stages of incorporation of tungsten into thefilm occur at locations beneath the pores and cell boundaries,respectively,due to the scalloping of the metal.Tungsten incorporated at the pore region should lie ahead of tungsten at the cell walls,due to outward migration of the ions,which is contrary to the real behaviour.Further,there is no evidence from TEM of a tungsten-rich layer at the base of pores that would be anticipated from preferential dissolu-tion of aluminium species if tungsten species were to reach the film/electrolyte interface.The distribution of tungsten species in the presentfilms is inverted with respect to expectations of a dissolution model,with species moving by ionic migration. Here,thefirst incorporated tungsten species,beneath the pores, eventually trail the species that are subsequently incorporated into the cell wall regions.There is no evidence for formation of anion species of tungsten following anodic oxidation of Al–W alloys to develop barrierfilms[17].Tungsten species have been found to migrate outward,whether incorporated from the alloy or from the electrolyte[17,18].Further,the outward movement of the tungsten species eventually located in the cell wall regions is consistent with the observations for barrierfilms.Formation of an anionic species in the region beneath the pore base would require a transformation of the tungsten species from a cation species to an anion species within thefilm.The consequent rever-sal of the direction of migration of tungsten species would lead to a separation of the distributions of cationic and anionic tungsten species,which is not observed by TEM.In contrast to the dissolution model,the presentfindings indi-cate that the constant thickness of the barrierfilm is maintained byflow of oxide from the barrier layer toward the cell wall, driven by compressive stresses from electrostriction[21]and possibly through volume expansion due to oxidation.Theflow is facilitated by the plasticity of the barrier layer due to partic-ipation of most of thefilm constituents in ionic transport[22]. Evidence offlow in anodic alumina is available from the ability of anodicfilms on Al–Cu alloys to accommodate the expan-sion of oxygen bubbles infilms,with bubbles having estimated gas pressures of the order100MPa[23,24].Similar stress lev-els have been estimated for electrostriction,which are sufficient to deform oxides[21].The compressive stresses act along the direction of thefield,the pressure facilitating theflow of anodic oxide around the scalloped metal/oxide interface.The voltage peak in the voltage–time response due to incorporation of tung-sten into the barrier layer is possibly aflow-related effect,since an opposite trend is expected from the formation ratios of anodic WO3,about1.7nm V−1[25],and Al2O3,about1.2nm V−1 [26],respectively,which suggest a reduced resistivity of the alumina.The displacement of the anodic oxide toward the cell walls is accommodated by the expansion of thefilm by a factor of about1.35compared with the thickness of the oxidized metal for the present conditions of anodizing.Other work onfilms formed in oxalic and sulphuric acid electrolytes,but using much thicker anodic oxides than those of the present experiments,indicates a similar expansion to that found here for a corresponding current density,with the expansion factor increasing at increased values of current density[27].The schematic diagrams of the tracer in thefilms formed for180and240s disclose the major distortion of the tracer band during its transit of the barrier layer(Fig.6(a and b)). Between these times,the metal/oxide interface recedes about 100nm,according to Faraday’s law,with formation of a similar thickness of oxide.The tungsten beneath the pores is moved outward by about30nm relative to the metal/oxide interface. Assuming a similar displacement of the tracer duringfilm growth for a further60s,no more than a few percent of the original tungsten should eventually reach the pore base(Fig.6(c)),which is in agreement with the negligible to low losses of tungsten species indicated by RBS.As evident from Fig.6,the tungsten remains within the barrier layer region while themetal/oxide Fig.6.Schematic diagrams showing the relative distributions of tungsten in anodicfilms at intervals of60s of anodizing at5mA cm−2in0.4M phosphoric acid solution at293K:(a)180(b)240and(c)300s.The distribution of(c) assumes a similar displacement of tungsten as in the previous60s.686S.J.Garcia-Vergara et al./Electrochimica Acta52(2006)681–687interface has retreated a distance of about twice the thicknessof the barrier layer.Fig.6(a)shows that the tungsten is almosthalf way though the barrier layer and according to a dissolutionmodel would reach the pore base if the interface retreats a furtherdistance of roughly about one-half the thickness of the barrierlayer,neglecting influences of tungsten migration.Thefigurestherefore suggest that the tungsten residency in the barrier layerhas been extended,due to theflow offilm material,by about thetime required for the metal/oxide interface to retreat a distanceof up to about1–1.5times the thickness of the barrier layer.The time required for the metal/oxide interface to retreat theprevious distance is about60–90s,although resolution of anyinfluence of tungsten in the barrier layer on the voltage–timeresponse at relatively long times will be hindered by the lowamount of tungsten in the high-field region.However,furtherstudy is required of theflow distribution and ionic migration inthe barrier layer in order to understand the precise behaviour ofthe tungsten species in the barrier region.It is of interest that oxygen tracer studies of porousfilmgrowth in sulphuric acid solution,in which afilm formed in 18O-rich electrolyte was further thickened by re-anodizing in16O-rich electrolyte,have revealed retention of practically all the original18O tracer in thefinalfilm[11].The behaviour was interpreted in terms offield-assisted dissolution but with re-incorporation of the18O tracer,released by dissolution,into film formed in the16O-rich electrolyte.However,suchfindings are also consistent with theflow model offilm growth,involv-ing onlyfield-assisted ejection of Al3+ions.According to the flow model,the18O tracer in the barrier layer beneath the pores is re-located to the cell walls during anodizing in the O16-rich electrolyte.Notably,tungsten-rich oxide is also found along the cellboundary regions of the presentfilms for distances up to200nmfrom the main tracer band indicative of either delayed incor-poration of tungsten into thefilm or re-location of previouslyincorporated tungsten species.Tungsten may be retained in themetal and then incorporated into thefilm over a protracted timedue to enrichment processes of tungsten residues[17],althoughrequiring transport within the enriched layer to the region of thecell boundaries.Re-location may occur by physical displace-ment offilm material or inward migration,the latter requiringan unlikely transformation to an anion species.4.2.Conditions for porousfilm growthThe conditions for porousfilm growth have been summarizedrecently by the authors[28].Barrier-type anodic aluminafilmsgrow by migration of Al3+ions outward and O2−ions inward[29],with respective transport numbers of about0.4and0.6.For growth at100%efficiency,about40%of thefilm thick-ness forms at thefilm/electrolyte interface,due to migration ofAl3+ions,the remainder forming at the metal/film interface bymigration of O2−ions.Porousfilm growth occurs when nofilmmaterial is added at thefilm/electrolyte interface,which corre-sponds to an efficiency of about60%for conditions close tothose presently used for anodizing[30].Aflatfilm/electrolyteinterface is then unstable in response to local perturbations of the electricfield that can stabilize embryo pores.Pore-filling is prevented by the absence of growth of new oxide at thefilm surface,while increased stresses from electrostriction assist sta-bilization of the pores.The sites of the initial pores depend upon the topography of the original aluminium surface[31].Some incipient pores stop growing,while others develop into the major pores of thefilm.4.3.Behaviour of incorporated anionsThe distributions withinfilms of characteristic species orig-inating from the electrolyte anions are reasonably known from past work[32].For porous alumina formed in sulphuric and phosphoric acids solutions,sulphur and phosphorus species are found in the barrier layer and cell wall,to relative depths of about0.95and0.7,respectively.The phosphorus and sulphur species migrate inward.Only inward migrating species can be found in thefilm material formed at the metal/oxide interface by inward migration of O2−ions,and hence found in porousfilms [33].In general,the incorporated anion species migrate more slowly than the O2−ions,leading to a layer of relatively pure alumina adjacent to the metal/oxide interface.However,accord-ing to the dissolution model,with inward migration of anion species,a non-thickening barrier layer should contain no incor-porated anion species,sincefilm growth proceeds at the rate at which O2−ions are transported to the metal/oxide interface, while anions move more slowly toward the interface.Thus,if anions were initially present within the outer region of the bar-rierfilm,the relatively pure alumina layer should increase at the expense of the anion-containing region.The revised model of pore generation is compatible with incorporation and inward migration of electrolyte species,since there is no significant field-assisted dissolution of the alumina that would eliminate their presence.The anion constituents may have key influences on theflow of material within the barrier layer,thereby affecting cell and pore dimensions.Such influences may arise from the rel-atively large size of the incorporated anions,and their effect on bonding of species within thefilm material.However,the trans-port offilm species within the barrier regions of porousfilms requires much more extensive studies,sincefindings for barrier films may be modified in the altered geometry of porous anodic films.Notably,the observations of relatively high expansion fac-tors,to1.65[27],with increased current densities,suggest the possibility of significantly altered processes to those encoun-tered in barrierfilms.Such aspects offilm growth are being addressed in the future work of the authors.5.ConclusionsPore generation in anodic aluminafilms formed in phosphoric acid solution is mainly a consequence offlow offilm material in the barrier layer region beneath the porous layer.The material flows from the region of pore bases towards the cell wall regions due to growth stresses andfield-assisted plasticity of thefilm material.It leads to an increased thickness of the anodicfilm, relative to that of the metal oxidized,by a factor of about1.35 for the selected conditions of anodizing.S.J.Garcia-Vergara et al./Electrochimica Acta52(2006)681–687687AcknowledgementsThe authors are grateful to the Engineering and Physical Sci-ences Research Council(U.K.)for support of this work.They also wish to thank Dr.I.Vickridge of the Group de Physique des Solides,Universit´e s Paris7et6,for provision of time on the Van de Graaff accelerator.References[1]S.Wernick,R.Pinner,P.G.Sheasby,The Surface Treatment and Finishingof Aluminium and its Alloys,Finishing Publications Limited,Teddington, 1996.[2]R.B.Mason,J.Electrochem.Soc.102(1955)671.[3]R.C.Plumb,J.Electrochem.Soc.105(1958)498.[4]F.Keller,M.S.Hunter,D.L.Robinson,J.Electrochem.Soc.100(1953)411.[5]J.P.O’Sullivan,G.C.Wood,Proc.R.Soc.Lond.317(1970)511.[6]H.Masuda,K.Fukada,Science268(1995)1466.[7]H.Masuda,H.Yamada,M.Satoh,H.Asoh,M.Nakao,T.Tamamura,Appl.Phys.Lett.71(1997)2770.[8]F.Li,L.Zhang,R.M.Metzger,Chem.Mater.10(1998)2470.[9]K.Nielsch,J.Choi,K.Schwirn,R.B.Wehrspohn,U.G¨o sele,Nano Lett.2(2002)677.[10]T.P.Hoar,N.F.Mott,J.Phys.Chem.Solids9(1959)97.[11]J.Siejka,C.Ortega,J.Electrochem.Soc.124(1977)883.[12]C.Cherki,J.Siejka,J.Electrochem.Soc.120(1973)784.[13]S.Ono,M.Saito,H.Asoh,Electrochim.Acta51(2005)827.[14]S.J.Garcia-Vergara,L.Iglesias-Rubianes,C.E.Blanco-Pinzon,P.Skeldon,G.E Thompson,P.Campestrini,Proc.R.Soc.A,in press.[15]G.Amsel,D.Samuel,Anal.Chem.39(1967)1689.[16]P.Skeldon,K.Shimizu,G.E.Thompson,G.C.Wood,Surf.Interface Anal.5(1983)247.[17]H.Habazaki,K.Shimizu,P.Skeldon,G.E.Thompson,G.C.Wood,J.Elec-trochem.Soc.143(1996)2465.[18]P.Skeldon,M.Skeldon,G.E.Thompson,G.C.Wood,Phil.Mag.B60(1989)513.[19]L.Iglesias-Rubianes,P.Skeldon,G.E.Thompson,H.Habazaki,K.Shimizu,Corros.Sci.43(2001)2217.[20]H.Habazaki,K.Shimizu,P.Skeldon,G.E.Thompson,G.C.Wood,X.Zhou,Trans.Inst.Met.Finish.75(1997)18.[21]N.Sato,Electrochim.Acta16(1971)1683.[22]J.P.S.Pringle,Electrochim.Acta25(1980)1423.[23]P.Skeldon,G.E.Thompson,G.C.Wood,X.Zhou,H.Habazaki,K.Shimizu,Phil.Mag.A76(1997)729.[24]X.Zhou,G.E.Thompson,M.A.Paez,P.Skeldon,H.Habazaki,K.Shimizu,G.C.Wood,J.Electrochem.Soc.147(2000)1747.[25]K.Shimizu,G.M.Brown,H.Habazaki,K.Kobayashi,P.Skeldon,G.E.Thompson,G.C.Wood,Corros.Sci.40(1998)1229.[26]A.C.Harkness,L.Young,Can.J.Chem.44(1966)2409.[27]I.Vrublevsky,V.Parkoun,V.Sokol,J.Schreckenbach,G.Marx,Appl.Surf.Sci.222(2004)215.[28]P.Skeldon,G.E.Thompson,S.J.Garcia-Vergara,L.Iglesias-Rubianes,C.E.Blanco-Pinzon,Electrochem.Solid State Lett.,in press.[29]F.Brown,W.D.Mackintosh,J.Electrochem.Soc.120(1973)1096.[30]G.E.Thompson,Y.Xu,P.Skeldon,K.Shimizu,S.H.Han,G.C.Wood,Phil.Mag.B55(1987)651.[31]G.E.Thompson,R.C.Furneaux,G.C.Wood,J.A.Richardson,J.S.Goode,Nature272(1978)433.[32]G.E.Thompson,G.C.Wood,Nature290(1981)230.[33]G.C.Wood,P.Skeldon,G.E.Thompson,K.Shimizu,J.Electrochem.Soc.143(1996)74.。

Analysis of Cetirizine Hydrochloride Using a Core Enhanced Technology Accucore HPLC ColumnJoanne Gartland, Thermo Fisher Scientific, Runcorn, Cheshire, UKAbstractThis application note demonstrates the use of the ThermoScientific Accucore HILIC HPLC column for the fastanalysis of cetirizine hydrochloride. The method of analysiscan be used as an alternative to the USP monograph whichuses an aggressive acid in the mobile phase.IntroductionAccucore™ HPLC columns use Core EnhancedTechnology to facilitate fast and high efficiencyseparations. The 2.6 µm diameter particles are not totallyporous, but rather have a solid core and a porous outerlayer. The optimised phase bonding creates a series of highcoverage, robust phases. The tightly controlled 2.6 µmdiameter of Accucore particles provides much lowerbackpressures than typically seen with sub-2 µm materials.Analyte properties that govern retention with Accucore HILIC are acidity/basicity, which determines hydrogen bonding, and polarizability which determines dipole-dipole interactions.Cetirizine is an antihistamine, used commonly for the treatment of allergies and hay fever.The USP uses an aggressive mobile phase containing sulphuric acid. We have demonstrated a similar separation using an alternative, non-aggressive buffer system.Sample PreparationWorking standard contained 50 µg/mL of cetirizine hydrochloride in 5:95water/acetonitrile (v/v)Thermo Scientific Column Part Number Accucore HILIC 2.6 µm 50 x 2.1 mm17526-052130 Measured pressure: 50 barThermo Scientific Accela HPLC systemColumn temperature30 °CInjection volume 1.0 µLFlow rate0.4 mL/minUV detection230 nmMobile Phase90:10 MeCN / ammonium acetate 200 mM pH5.0Consumables Part NumberFisher Scientific HPLC grade water W/0106/17Fisher Scientific HPLC grade acetonitrile A/0626/17NSC Mass Spec Certified 2 mL clear vial with blue MSCERT4000-34W bonded PTFE silicone capPart of Thermo Fisher ScientificIn addition to these offices, Thermo Fisher Scientific maintains a network of represen -tative organizations throughout the world.North America USA and Canada +1 800 332 3331Europe France+33 (0)1 60 92 48 34Germany+49 (0) 2423 9431 -20or -21Switzerland +41 56 618 41 11United Kingdom +44 1928 534110Asia Japan+81 3 5826 1615China+86-21-68654588or +86-10-84193588800-810-5118India1800 22 8374 (toll-free)+91 22 6716 2200Thermo FisherScientific Australia Pty Ltd1300 735 292(free call domestic)Thermo Fisher Scientific New Zealand Ltd0800 933 966 (free call domestic)All Other Enquiries +44 (0) 1928 534 050Technical SupportNorth America 800 332 3331Outside North America+44 (0) 1928 534 440/chromatography©2011 Thermo Fisher Scientific Inc. All rights reserved. All trademarks are the property of Thermo Fisher Scientific Inc. and its subsidiaries.Specifications, terms and pricing are subject to change. Not all products are available in all countries. Please consult your local sales representative for details.ANCCSCETCHYDCHL 0611ResultsThe original USP analytical conditions were based on a L3250 x 4.0 mm, 5 µm column using a mobile phase of acetonitrile, water and 1M sulphuric acid (93:6.6:0.4).Sulphuric acid is an aggressive acid which can damage steel HPLC components. Typical run times for the USP application are approximately 11 minutes.The analysis was carried out on an Accucore HILIC 2.6 µm 50 x 2.1 mm column. Cetirizine is eluted in less than 1 minute (Figure 1), which demonstrates over a 10-fold reduction in analysis time in comparison to the original method. The USP acceptance criteria (Tailing factor <2.0, %RSD t r <2.0 and %RSD peak area <2.0)were achieved (Table 1.) The statistical assessment is based on data from 6 replicate injections.ConclusionsThe use of Accucore HILIC column successfully retained cetirizine hydrochloride without the use of an aggressive acid in the mobile phase, which is used in the USPmethod. The analytical results exceeded the requirements stated in the USP monograph. Accucore HILIC columns are therefore an excellent choice for the fast analysis of cetirizine hydrochloride allowing high sample throughput.Figure 1: Chromatogram for cetirizine hydrochloride retained on an Accucore HILIC 2.6 µm 50 x 2.1 mm columnTable 1: Method precision (%RSD) for cetirizine hydrochloride (data calculated from six replicate injections)Time (Minutes)0.00.20.40.60.8 1.0 1.2 1.41.61.82.0m A U-20 0 20 40 60 80 100 120 140 160 USP Specifications Cetirizine HydrochlorideTailing factor <2.0 1.21%RSD t r <2.0% 1.39%RSD area <2.0%0.17。

Design,Optimization and Control of Extractive Distillation for the Separation of Trimethyl Borate −MethanolZhenyu Bao,Weijiang Zhang,Xianbao Cui,and Jiao Xu *Department of Chemical Engineering,Tianjin University,Tianjin 300072,China1.INTRODUCTIONTrimethyl borate (B(OCH 3)3)is an essential reagent for the production of organic boride,sodium borohydride,high-purity boron,antifriction additive of lubricants,stabilizer and plasticizer of polymers,etc.Due to its particular physical and chemical properties,trimethyl borate is also used as a catalyst,dehydrant,auxiliary solvent in brass-welding and high-energy fuel in aircrafts.Natural abundance boron contains 19.8%B-10and 80.2%B-11,1which are the only two stable isotopes.Natural abundance trimethyl borate is normally synthesized by esteri fication of boric acid and methanol.It can also be synthesized by the reaction of borax,sulfuric acid and methanol.The reaction mixtures of the two synthesis methods contain trimethyl borate and methanol,which should be separated for further utilization.As B-10has excellent neutron absorption performance,it is widely used in nuclear power plants,military equipment and medical treatment.High abundance boron used in pressurized water reactors often exists in the form of boric acid.2Production of high B-10abundance boric acid is reported by Han et al.,3which inevitably forms a mixture of trimethyl borate and methanol,and the yield of target product is strongly a ffected by the separation of the mixture.At 101.3kPa,trimethyl borate and methanol form a minimum-boiling azeotrope,the azeotropic point is 328.37K and its azeotropic mixture contains 77−78%(wt %)trimethyl borate.4The azeotrope cannot be separated by ordinary distillation,so special distillation such as extractive distillation and other separation methods are introduced to separate the azeotropic mixture.Sulfuric acid dealcoholization is one of the industrially used methods.The product is an intermediate containing 92%(wt %)trimethyl borate.Then,a salting-out process is employed to get higher purity trimethyl borate (i.e.,98%(wt %)).Usually,lithium chloride is applied.5The separation process consumes a lot of sulfuric acid,which is corrosive,and the salting-out process is inconvenient to carry out.Extractive distillation is an important technique to separate close-boiling point mixture and azeotropes.It is based on the preferential a ffinity of solvent (entrainer)for one or more of the components over the others,which thus alters the relativevolatilities of the feed components.Khoury has given detailed descriptions on such a process.A lot of azeotropes have been separated via extractive distillation;however,up to now,the separation of the azeotrope of trimethyl borate −methanol by extractive distillation has not been reported in open literature.Solvent selection is a major task in extractive distillation.In recent years,many methods have been developed to select proper solvents,such as qualitative judgment,quantitative estimation and experimental determination.Pretel et al.7and Papadopoulos and Linke 8have applied computer-aided molecular design (CAMD)in single solvent design.Moreover,Karunanithi and Achenie 9have studied solvent mixture designs.Dyk and Nieuwoudt 10,11have extended genetic algorithm based CAMD models to design solvents mainly used for the distillation process.Conceptual design and optimization of distillation process have been investigated by Douglas,12Doherty and Malone 13andRodri guez-Donis et al.14Their research e fforts have shown how to select solvents and optimize column sequences.Because dynamic performance proves to be a necessary way to assess the e ffectiveness and controllability of steady state design,Sakizlis 15has presented reviews of integrated design and control methodologies.Shirsat et al.16and Ghaee et al.17have given detailed cases concerning extractive distillation.An e ffective plantwide process control system includes several goals such as safe and smooth process operation and tight control of product quality in front of disturbances.Luyben et al.18−22have presented essential tips and theories on dynamic control using Aspen Plus Dynamics and HYSYS.In their works,development of rigorous simulation of single distillation columns and sequences of columns are concluded;the interaction between steady state design and control is explained;plantwide control with emphasis on selection of control structures for an entire multiunit process is introduced;feedforward,feedback and protective controls areReceived:May 19,2014Revised:August 18,2014Accepted:August 26,2014applied to achieve automatic startup,shutdown and smooth,noninteracting control of column product composition.To fill in the gap of research on trimethyl borate −methanol separation technique,in this study,a proper solvent was selected with a combination of software calculation and experiment determination,and the solvent was comprehensively compared with other potential solvents.Then,the process was optimized through an iteration algorithm.Optimized parameters were further testi fied using the single factor method.At last,control structures were proposed to improve the dynamic stability of the system.2.PROCESS DESIGN AND OPTIMIZATION2.1.Solvent Selection.Solvent selection is the key problem for extractive distillation.The performance of a solvent can be indicated by several indicators based on the activity coe fficient at in finite dilution (γ∞).The commonly used indicators are solvent selectivity,solvent power,relative volatility at in finite dilution in the solvent and performance index.Their de finitions are as follows:9,7βγγ=∞∞solvent selectivity1,S 2,S(1)γ=·∞solvent power SP 1MW MW 2,S 2S(2)αγγ=∞∞P P relative volatility1,21,S 1S2,S2S (3)α=x performance indexPI MW 1,2S ms(4)where x ms is the minimum solvent molar fraction to break theazeotrope.For estimation of γ∞,an activity-coe fficient estimation model called conductor-like screening model for segment activity coe fficient (COSMO-SAC)is used,which is based on the interactions between surface charge distributions of molecules in solution.23Molecules are first transferred from vacuum phase to an ideal conductor,where surface charge is ideally shielded,then transferred to real solvent using polarity factors,rather than directly into condensed phase.COSMO-SAC obtains the molecular segments activity coe fficient using solvation free energy,which avoids the violation of thermodynamic consistency principle in several boundary conditions.24−26Most approaches in selecting solvents are mainly based on the same concept of molecular generation,testing and matching the molecules with speci fied target molecular properties.27−30Considering the distinct polarity di fference between trimethyl borate and methanol,combined with price,toxicity and stabilityconsideration in industrial application,several frequently used strong polar solvents such as dimethyl sulfoxide (DMSO),N ,N -dimethylformamide (DMF),N ,N -dimethylacetamide (DMAC),ethylene glycol (EG),glycerin and N -methyl-2-pyrrolidinone (NMP)were investigated.For each solvent,ten vapor −liquid equilibrium (VLE)data points were experimentally determined using the same method and facilities with Tian et al.32The regression and correlation were made using ChemCAD.The minimum solvent molar fraction (x ms )to break the azeotrope was then obtained,and the result is shown in Table 1.Meanwhile,vapor pressures of trimethyl borate and methanol were calculated using the Antoine equation with coe fficients obtained from literature.33,34As can be seen from Table 1,DMSO has comparatively higher β·SP and PI values than other solvents,which indicates a favorable e ffect on separation of the azeotrope.Moreover,it has a relatively low heat capacity,so it is more energy saving than other solvents.There is an interesting phenomenon that the x ms value is lower for NMP than for DMSO.But with increasing the solvent molar fraction,the solvent e ffect of DMSO increases sharply,which displays a rather high relative volatility when the solvent molar fraction is 0.4(see Figure 1).This obvious tendency was validated by further increase of the DMSO molar fraction.Figure 1shows the XY phase diagram for trimethyl borate −methanol with di fferent solvents (solvent molar fraction is 0.4)at 101.3kPa.Values and curves of EG and glycerin are not shown in Table 1and Figure 1because their miscibility and flowability are bad;moreover,the problem can be aggravated when mixing with trimethyl borate and methanol,as the temperatures of the mixtures are reduced.Table 1.Results of Solvent Selectionsolvent T b (K)γ1,S ∞γ2,S ∞βSP β·SP C P a (J ·mol −1·K −1)x ms α1,2100×PI DMSO 462.15 1.26850.3925 3.2323 1.0449 3.3774149.390.33 1.13 4.37DMF 426.150.98740.5161 1.91330.8494 1.6252146.050.440.87 2.72DMAC 438.150.97060.3478 2.7906 1.0574 2.9508178.20.361.163.71EG 470.35 3.09510.9719 3.18460.5311 1.6913149.8glycerol 563.15 2.89170.9571 3.02130.3635 1.0982218.9NMP476.150.91810.30473.01291.06083.1961412.40.290.95 3.30aData from NIST Chemistry WebBook.31Figure 1.Isobaric VLE diagram for the system of trimethyl borate (1)+methanol (2)+solvent (3)(x 3=0.4)at 101.3kPa.In Figure 1,x 1′represents the mole fraction of trimethyl borate in the liquid phase excluding the solvent,y 1is mole fraction of trimethyl borate in the vapor phase.It can be seen that DMSO can greatly enhance the relative volatility of trimethyl borate and methanol when the molar fraction is 0.4.NMP is another potential solvent with a strong solvent e ffect,especially when separating mixtures with high trimethyl borate concentrations.Whether the mixture of DMSO and NMP can produce better performance needs further investigation.In contrast,DMAC and DMF seem to be not as good as the former two solvents.The overall order of the solvent performance may be resulted from group interaction di fference between the mixture and the N-containing/S-containing compounds.2.2.Feasibility Study.The feasibility study was carried out by Aspen Plus V7.2.The UNIQUAC activity coe fficient model was used to calculate the vapor −liquid equilibrium;however,only the binary-interaction parameters of methanol −DMSO were found in the database.The binary-interaction parameters of trimethyl borate −methanol were obtained from Gmehling ’s handbook,35and the binary-interaction parameters of trimethyl borate −DMSO were correlated from the vapor −liquid equili-brium data calculated by COSMO-SAC.Results are shown in Table 2,and the parameters of UNIQUAC model are 0except for bij and bji.A residue curve map of the ternary system was obtained through Aspen Plus and is shown in Figure 2.As there exists noboundary line in the map,the azeotrope and DMSO is the unstable and stable node,respectively,so the extractive distillation process is feasible.In this figure,black bold lines are the material balance line,F stands for the composition of the material to be separated (fresh feed).F1is the mixture of fresh feed and solvent,which can be separated into D1(overhead of extractive distillation column)and B1(bottom product of extractive distillation column),then B1can be separated into D2(overhead of solvent recovery column)and B2(bottom productof solvent recovery column).D1and D2are trimethyl borate and methanol products,respectively,so it is feasible to separate the azeotrope into pure products with the aid of DMSO.The dashed line across the map is the isovolatility curve.36When DMSO is added,the isovolatility line moves toward the hypotenuse (where the concentration of methanol is zero).Meanwhile,because DMSO lies in the part below the isovolatility curve,where the relative volatility α(C 3H 9BO 3/CH 4O)>1,causing trimethyl borate to go up the column.So,trimethyl borate is the main product of D1.2.3.Partial Optimization.The feed to be separated contains 77%(wt %)trimethyl borate and 23%(wt %)methanol,and the flow rate is 3000kg/h.The product speci fications are the following:D1,99.5%(wt %)trimethyl borate;D2,99.5%(wt %)methanol.The nonoptimized flowsheet is shown in Figure 3.In Figure 3,the variables with interrogation mark behind should be optimized,which will be discussed in the following part.A traditional separation sequence is adopted.As the solvent is recycled for the utmost utilization,a solvent makeup flow is necessarily added to balance the tiny solvent loss.2.3.1.Design of Extractive Column (C1).Because the price of trimethyl borate is much higher than that of methanol,trimethyl borate should be recovered as much as possible.Speci fication for the bottom product is the following:trimethyl borate flow rate should be no more than 0.5kg/h (recovery rate:99.98%).There are five variables in C1that need to be optimized:total stages (N T1);fresh feed stage (N F1);solvent feed stage (N FS );solvent flow rate (S );re flux ratio (RR1).At first,we set the total stages and the feed stage of solvent recovery column at 16and 6,respectively.N T1is given as 62;after trial and error,the minimum energy consumption value (QR1+QR2)is obtained when N F1is set at 48and N FS at 4.Meanwhile,N F1and N FS values other than those always result in disquali fication of products.Similarly,to reach the speci fications,N T1should be no less than 62.That preliminarily con firms the minimum N T1and the proper N F1.Then,concentrations of components in the overhead flow of the extractive column at di fferent S are studied by changing RR1.The results have been plotted in Figure 4,in which the results are obtained when bottom speci fication is satis fied.From Figure 4,we can see that the trimethyl borate concentration reaches a maximum value when RR1=0.4;at the same time,the concentration of impurities is almost the least.Table 2.UNIQUAC Model Parameters of the Ternary Systemcomp,i C 3H 9BO 3C 3H 9BO 3CH 4O comp,j CH 4O DMSO DMSO bij −629.79−162.60129.36bji76.44100.6623.49Figure 2.Residue curve map for trimethyl borate −methanol −DMSO (101.3kPa).Figure 3.Non-optimized flowsheet for the process.Meanwhile,an increase of S results in a higher product purity.However,more solvent also leads to a larger QR1+QR2.The green lines in Figure 4are the purity requirements.As we have de fined that product purity should be 99.5%(wt %),S =24000kg/h is adopted.If we change the purity requirement to 99%(wt %),S can be lowered to 18000kg/h,which can save a lot of energy.Adjusting the operation condition to meet the corresponding purity requirement is suggested when implement-ing the project.The concentration of DMSO is constant when changing S ,because the speci fication is used,but it can be seen that solvent loss reaches the minimum when RR1=0.4.2.3.2.Design of Solvent Recovery Column (C2).Separation of methanol and DMSO takes place in C2.It is relatively easy due to a large relative volatility (76.07,calculated using Aspen Plus built-in data).Methanol is the second target product;99.5%(wt %)of D2is methanol.Most of DMSO enters B2with trace impurity inside (1ppm).There are three variables that need to be optimized:total stages (N T2);feed stage (N F2);re flux ratio (RR2).By fixing the obtained values for C1,we set N T2at 16,then N F2is tested to minimize the energy consumption.Stage 6is selected for N F2,and the initial RR2is estimated by using the same method in the C1design.N T2values other than 16and N F2values other than 6are discussed in the next part.As RR2is manipulated by Aspen Plus to attain the product speci fication,RR2=2.0is used as an initial assignment.2.4.Global Economic Optimization.The tradeo ffbetween economic bene fits,product quality and controllability has been studied by Brengel and Seider,37Luyben and Floudas,38and Palazoglu and Aarkun.39It is a complex job to take every factor into account,so in this study,only utility consumption is considered when calculating operating cost.Total annual cost (TAC)consists of fixed capital investment (FCI)and cost of utilities (C UT ).We will adopt the equation mentioned in Mun o z ’s research:40=+C TAC (10$/a)0.3FCI3UT (eq.5)By observing the proper RR values,we can see that when total stages reduces,FCI drops obviously;however,RR values need to increase to meet the speci fications,which,in turn,raise C UT .TAC is used here to search for the optimum condition.Global optimization is undertaken in the following iterative manner:(1)Fix the pressures at the top of the two columns at 101.3kPa.(2)Give initial estimates for N T2and N F2.(3)Give values for N T1,N F1and N FS .(4)Set S at the minimum value obtained in the partialoptimization part.(5)Change the values of S ,D1and RR1to achieve thespeci fications of C1.(6)Go back to (4)until QR1+QR2is minimized with N T2,N F2,N T1,N F1and N FS fixed.(7)Go back to (3)until TAC is minimized with N T2and N F2fixed.(8)Give values for N T2and N F2.(9)Change the values of D2and RR2to achieve thespeci fications of C2.(10)Go back to (9)until QR1+QR2is minimized with N T1,N F1,N FS ,N T2and N F2fixed.(11)Go back to (4)until TAC is minimized with N T1,N F1andN FS fixed.(12)Go back to (2)until TAC is totally minimized with onlypressures fixed.The optimization procedure is brie fly shown in Figure 5.The results have been listed in Table 3.In Table 3,D 1/2stands for the diameter of the columns.C UT includes the cost of cooling water (0.354$/GJ),steam (14.19$/GJ)and electricity (16.8$/GJ),which are taken from Turton ’s book.41Pumps are considered as electricity-consuming devices.Cost of facilities such as columns,condensers and reboilers are included in FCI,because these large instruments need annual maintenance.In section 2.3(partial optimization),N F1and N FS have been optimized with the minimum energy consumption when N T1=62.Because 62is veri fied to be the optimum value for N T1,then N F1=48and N FS =4are fixed in each case except case 1(N F1=50is optimal).For now,TAC values are calculated to compare the best con figuration of the other variables.Cases 1and 2showFigure 4.Concentration of components in D1at di fferent RR1and S values.how N T1affects the TAC.Herein,N T2and N F2arefixed with the optimum value.As can be seen,TAC is lower when N T1=62, which is also the minimum N T1that can achieve the specifications we set.Likewise,cases3,4and5show how N T2affects TAC.In these cases,the middle part of the columns are selected as N F2.It is clear that N T2=16is superior to N T2=18and14.Cases4,2, and6demonstrate the best N F2should be6.RR1is not the best value we estimate due to surplus purification that we defined.But it does not matter that we choose RR1=0.4,as operation condition,purer products can be obtained in that case.For smaller reflux ratios such as0.33,they can lead to a larger output rate,which is beneficial for a plant.2.5.Analysis of Optimization Result.The single factor method is used to test the validity of the optimization procedure. N T1,N F1,N FS,N T2and N F2are studied besides the above-stated RR and S values.Results are shown in Figure6.In Figure6a,N T1is found to be exactly suitable in achieving the product specification at62,and the purity of trimethyl borate increases with the increase of N T1.Figure6b,c shows that the optimal N F1and N FS values are achieved at48and4,respectively, which are found to be the only values satisfying the given specifications.These locations are influencing because they are either near the bottom or near the top of the column.Increasing N F1or decreasing N FS results in a sharp decrease in product purity.In comparison,a more gradual increase in product purity is observed in C2when N T2is larger than16,as shown in Figure 6d.Moreover,product purity stops increasing when N F2is larger than6,which is demonstrated by Figure6e.The different tendencies between the two columns occur due to the difference in separation difficulties.The more difficult to separate a mixture, the greater impact variables cause on product purity.In conclusion,purity of products can reach the specifications we expected with the optimized values.Diameters of the two columns denote the internal diameter, which is calculated by Aspen inner algorithm and rectified by eq 6:42ρ==F V1max V(6) where V max is the maximum vapor velocity(ft/s)andρV is the corresponding vapor density(lb/ft3).The maximum vapor volumetricflow rate andρV are inquired in hydraulic parameters, then V max is calculated by eq6.The cross-sectional area and internal diameter of the column are thus obtained.By comparing with the column diameter in tray sizing tab,tray spacing is adjusted to make them equal.The tray spacing is0.24m forC1 Figure5.Optimization procedure sequence.Table3.Global Optimization Resultparameters case1case2case3case4case5case6 N T1646262626262N T2161618161416N F1/N FS50/448/448/448/448/448/4 N F2669874RR10.30.330.330.330.330.33RR2 1.73 1.71 1.92 2.32 2.95 1.82D1(m) 1.563 1.565 1.565 1.565 1.565 1.565 D2(m) 1.010 1.011 1.035 1.072 1.137 1.023 Q C1(kW)−261.49−266.30−266.38−266.30−266.30−266.36 Q R1(kW)2449.812448.432448.502448.432448.432448.48 Q C2(kW)−573.41−569.78−614.17−697.23−830.00−593.29 Q R2(kW)805.15807.58859.17935.031060.02831.09C UT(103$)2606.252572.472626.772625.862645.612582.34FCI(103$)747.82723.02737.11729.76724.38724.17 TAC(103$)2830.592789.382847.902844.782862.932799.59and 0.253m for C2.This procedure can ensure the column height is correct,which is essential for economic evaluation.At last,the total solvent loss is calculated by material balance equation,thus the makeup flow rate is 4.077kg/h.A global flowsheet is presented in Figure 7with detailed stream and facility operation information in it.The optimal values of parameters shown in Figure 7are consistent with the case 2values in Table 3.It can be seen that solvent usage is economized,as the S /F ratio is less than 0.14%(wt %)when operating.Even so,the amount of recycled solvent is large,further that utilization of the tremendous energy that the recycled flow possessed is bene ficial.Temperature distribution along the columns and concen-tration of the three components in liquid phase are displayed in Figure 8.Figure 8gives a description on how separation occurs on each stage.For C1,the temperature rises rapidly in the above four stages and turns to be flat between stages 4and 47,then becomes steep again;this is because cold solvent and fresh material are fed on stages 4and 48,respectively.Three sections are dividedapparently and each one has a special function.We may as well label these three sections as s1,s2and s3from column top to bottom.As can be seen from colored lines in Figure 8a,DMSOFigure 6.Correlations between product purity and N T1(a),N F1(b),N FS (c),N T2(d)and N F2(e).Figure 7.Optimal design flowsheet.concentration drops sharply from stage 4to stage 1in s1,as high purity trimethyl borate is required in D1.s2serves to minimize methanol concentration in the overhead,but methanol concentration drops slowly from stage 48to stage 4,which is due to the di fficulty in azeotrope separation.Trimethyl borate is stripped o ffin s3,while methanol and DMSO enter C2.Similarly,but seeming simpler in C2,only a tiny amount of trimethyl borate appears between stages 2and 6,the primary task for this column is to split methanol and DMSO.3.CONTROL SYSTEM DESIGNTo achieve the control purpose,many control strategies with di fferent combination of manipulated variables con figurations have been proposed by Skogestad.43He has elaborated these strategies from theoretical points.More recently,Luyben 42has discussed a variety of control structures with the aid of computer software,one category of them is called “Single end ”control structure where the temperature on one tray is controlled.Moreover,based on Luyben ’s “slope criterion ”42in the determination of control point,temperature measurement is easy,fast and accurate,and composition is closely related with temperature,the stage that has a rapid temperature change (large slope in temperature pro file)can be chosen as an excellent control point.Because this temperature can be used as an obvious re flection of the key component composition variation.Maintaining the temperature of this tray can keep the composition pro file in the column unchanged and thus prevent impurities from entering the withdraw flow.Based on that,stage 61and stage 3are chosen for C1and C2,respectively.Temperature drops on the feed stages are also large,but it is not suitable for control,as the fluctuation of feed conditions can easily destroy the e ffectiveness of the control system.An e ffective and practical control structure needs a precise control point and accurate equipment dimensions.A commonly used heuristic is applied here:42the re flux drum and column base are supposed to provide 10min of liquid holdup,and the length-to-diameter ratio is 2:1.In this case,the re flux drum is 0.72and 0.62m in diameter for C1and C2,respectively.The column base is 1.55and 1.49m in diameter for C1and C2,respectively.3.1.Control Structure with Fixed RRs.As we have obtained RR1and RR2in the optimal operation condition;they are entered in the multipliers.Other manipulated variables and their control actions are listed as follows:(1)Feed flow rate is controlled by adjusting inlet valve(Controller:FC,reverse acting).(2)Solvent flow rate is controlled by adjusting V2B,which is 8times feed flow rate (controller:SFC,direct acting).(3)Temperatures of the certain stage in both columns arecontrolled by adjusting reboiler heat duties (controllers:TC1and TC2,reverse acting).(4)Pressures in both columns are controlled by adjusting thecondenser heat duties (controllers:PC1and PC2,reverse acting).(5)Temperature of recycled solvent is controlled by adjustingheat removal rate of the cooler (controller:TC3,reverse acting).(6)Re flux drum levels in both columns are controlled byadjusting withdraw flow rate of distillates (controllers:LC1D and LC2D,direct acting).(7)Base level in extractive distillation column is controlled byadjusting withdraw flow rate of B1(controller:LC1B,direct acting).(8)Base level in solvent recovery column is controlled byadjusting solvent makeup flow rate (controller:LC2B,reverse acting).In Figure 9,solvent flow rate is manipulated by V2B rather than VM because makeup flow is too small to handle fluctuates when operating.SFC is a cascade controller;it can precisely regulate V2B according to the ratio of SF output (8times feed flow rate)and total solvent flow rate.PID controllers are used here except temperature controllers and pressure controllers.Level controllers are set at K C =2and integral time =9999min,because a proportional-only control is used.Flow controllers use the conventional tuning:K C =0.5,integral time =0.3min.Pressure controllers use default settings.Temperature controllers are PIDincr controllers,and they are tuned with dead time =1min.A Tyreus −Luyben tuning rule is selected to update calculated gains and integral times.Close loop tuning is used with relay amplitude of 5%output range.Detailed tuning parameters are listed in Table 4.Dynamic performance is tested by feed flow rate and feed composition disturbances when time is 0.1min.However,this control structure is not so e ffective to deal with several disturbances.It works well for +20%feed flow rate (3000kg/h →3600kg/h),but it needs 20h to keep the system steady if the feed flow rate decreases 1%(3000kg/h →2970kg/h).For composition disturbances,integrator fails to work when trimethyl borate concentration in the feed changes 1%(both positive and negative).The results for feed flow rate disturbances are shown in Figure 10.It can be seen that variables are brought to new steady state in 3h.New temperatures for both columns are consistent with original ones,but the purity of products is changed.More pronounced is methanol concentration changes in D2whenaFigure 8.Temperature and liquid concentration pro files.。

DATASHEET Overview Avalon software system is the next-generation CAD navigation standard for failure analysis, design debug and low-yield analysis. Avalon is a power packed product with tools, features, options and networking capability that provides a complete system for fast, efficient and accurate investigation of inspection, test and analysis jobs. Avalon optimizes the equipment and personnel resources of design and semiconductor failure analysis (FA) labs by providing an easy-to-use software interface and navigation capabilities for almost every type of test and analytical failure analysis equipment.Avalon enables closer collaboration of product and design groups with FA labs, dramatically improving time to yield and market. Avalon can import CAD design data from all key design tools and several user-proprietary formats while providing visual representations of circuits that can be annotated, exploded, searched and linked with ease.Benefits • Improves failure analysis productivity through a common software platform for various FA equipment • Significantly decreases time to market with reduced FA cycle time • Faster problem solving by cross-mapping between device nodes to view all three design domains (layout, netlist and schematic) simultaneously • Increases accuracy of FA root cause analysis using advanced debug tools • Single application that overlays images from various FA equipment on to design layout • Secure access to all FA information using KDB™ database • Design independent system that supports all major layout versus schematic (LVS)• Complete access to all debug tools critical to failure trace, circuit debug and killer defect source analysis • Simple deployment setup with support for Linux and Windows • Seamless integration with legacy Camelot™ and Merlin™ databases • Ease of conversion for layout, netlist and schematic data and establishes cross-mapping links between each data entityCAD Navigation andDebug Solutions forFailure AnalysisAvalonFigure 1: Avalon CAD-navigation system integrating layout, signal tracing and 3D viewSupporting all CAD Design DataSynopsys is committed to being the leading provider of software solutions that links all CAD design data. Avalon is a comprehensive package that reads all EDA tools and design data from verification systems and several user-proprietary formats. The KDB™database is designed to interface with all key design formats.Today, there are more EDA developers and more verification package choices; Synopsys is the only company thatsupports all of them.• LVS Conversions: Cadence (Assura, DIVA), Mentor Graphics (CheckMate, Calibre), Synopsys (Hercules, ICV)• Netlist Conversion: SPICE, EDIF, OpenAccess• Layout Conversion: GDSII, OASIS®The highest priorities for Avalon users are faster data accessibility, support diverse failure analysis equipment and availability of debug tools. Avalon provides the optimal solution for both small and continually-expanding FA labs and design debug teams. The Avalon database is design independent and offers a superior level of data consistency and security. The unique design of the internal database schema guarantees compatibility with decades-old databases. This is an indispensable feature for all failure analysis, QAand manufacturing organizations especially in the automotive industry.Figure 2: Avalon SchemView and NetView provide an easy way to navigate inside circuit schematicsProviding Critical Analysis FunctionsIn addition to its CAD navigation and database capabilities, Avalon’s analysis features have become indispensable to the FA lab. Different viewing options are critical in tracking potential failures and determining the source and origin of killer defects. Avalon includes special schematic capabilities and layout features that are invaluable to FA engineers as they debug chips manufactured using new processes.Avalon View Only Client consists of maskview, netview, schemview, i-schemview, K-EDIT, defect wafermap and 3D-SAA. The list below details some of the most commonly used applications.Defect Wafer Map integrates defect inspection data with the device CAD design using the defect coordinates to navigate an equipment stage and pinpoint the defect for closer inspection and characterization. Avalon sorts defects by size, location or class, as well as layout location and allows the user to define custom wafer maps. Additionally, users can classify defects, attach images and write updated information to the defect files.Figure 3: Defect Wafer Map pinpoints defects for closer inspectionSchemView provides tracking of potential failures through visualization of the chip logic. Cross-mapping of nets and instances to the device layout and netlist, SchemView helps determine the source and origin of chip failures. SchemView helps determine the source and origin of chip failures. The entire design is displayed in cell hierarchy format, allowing push-down to a transistor level.Figure 4: K-Edit allows collaboration between design, fab and labI-Schem (Interactive Schematic) creates a schematic from a netlist in a net-oriented format allowing forward and backward tracking to locate a fault. Features like Add Driver or Add Input Cone allow for quick analysis and verification of diagnostic resultsin scan chains.Figure 5: I-Schem creates a schematic from a netlistK-Bitmap allows equipment CAD navigation when analyzing memory chips by identifying the physical location of failingmemory cells. It eliminates tedious screen counting by converting the logical addresses, or row and column coordinates, to thephysical location.Figure 6: K-Bitmap identifies the physical location of bit addresses in memory devices3D Small-Area Analysis provides a three-dimensional cross- section capability to FA engineers, enabling faster localization of circuit failures to accelerate IC manufacturing yield improvement.Figure 7: 3D Small-Area Analysis enables faster localization of circuit failuresHot-Spot Analyzer allows user to draw regions on the layout that correspond to hot-spot regions (emission spots) to detect the crucial nets. It finds the nets in each hot-spot region and plots a pareto graph of nets crossing one or more hotspots which helps to easily locate the killer net.Figure 8: Hot-Spot Analyzer displays number of nets in a hot spotUser-Defined Online Search (UDOS) allows users to search a small area of a die for unique polygon features, repeated features or lack of features. Applications include, but are not limited to, FIB-able regions, repeaters, pattern fidelity and lithographic applications.Figure 9: User-Defined Online Search (UDOS) finds easy-to-access tracesPassive Voltage Contrast Checker (PVC) quickly and accurately validates the integrity of a circuit’s conductivity and provides detailed information for identifying suspect faults at via or metal tracesFigure 10: Passive Voltage Contrast (PVC) Checker identifies suspect vias or metal tracesElectronic Virtual Layer marks objects to represent net connectivity during a FIB deposit or cut using KEdit. The online trace will simulate the new connectivity to the virtual layer. PVC checker could be used on this virtual layer to simulate the crack or short.Check Adjacent Nets allows logical analysis of nets. This command line tool finds the adjacent nets which are within user-specified threshold distance to find shorts.Export Partial Layout enables the customer to share partial layout data with service labs without compromising the IP of the product.Image Mapper automates the image alignment process in Avalon Maskview and saves a lot of time and effort spent inmanual alignment.Advanced 3D Viewer displays real time 3D view of the selected layout area. It shows each process step in the 3D view for which it uses the process data along with design data. It zooms into smaller details and helps to minimize unintended consequences during FIB cuts due to underneath high density structure.Avalon SolutionAvalon brings all the advantages of enterprise-wide computing for FA of the chip. Avalon is an open architecture system that connects users over local and wide area networks for seamless integration and database sharing. Instrument integration throughout the fab and other locations throughout the enterprise enables viewing, modifying, characterizing and testing the same wafer location with different instruments, or the same location on wafers at different facilities using the same chip design.Figure 11: Avalon’s open architecture integrates with Synopsys’ Yield ExplorerIC DesignToolsFigure 12: Avalon server solutionComprehensive Library of FA Tool DriversAvalon provides navigation with almost every equipment used in the FA lab. With a continued commitment to support drivers for all types of test and analysis equipment, Synopsys will continue to develop driver interfaces for new tools as they are introduced to the market, as well as the next generation of existing tools.Equipment Supported by Avalon• Analytical Probe Stations• Atomic Force Microscopes• E-Beam Probers• IR Imaging• Mechanical Stage Controllers• Emission Microscopes• Microanalysis Systems• FIB Workstation• Laser Voltage Probe• LSM• EDA LVS• Microchemical Lasers• OBIC Instruments• Optical Review• SEM Tools• Photon Emission Microscopes• Laser Scan Microscopes©2018 Synopsys, Inc. All rights reserved. Synopsys is a trademark of Synopsys, Inc. in the United States and other countries. A list of Synopsys trademarks isavailable at /copyright.html . All other names mentioned herein are trademarks or registered trademarks of their respective owners.。

a r X i v :c o n d -m a t /0305550v 1 [c o n d -m a t .d i s -n n ] 23 M a y 2003A realization of Yangian and its applications to the bi-spin systemin an external magnetic fieldShuo Jin a,b,1,Kang Xue c ,Bing-Hao Xie a,ba Theoretical Physics Division,Nankai Institute of Mathematics,Nankai University,Tianjin 300071,P.R.china b Liuhui Center for Applied Mathematics,Nankai University and Tianjin University,Tianjin 300071,P.R.Chinac Physics Department,Northeast Normal University,Changchun,Jilin 130024,P.R.chinaAbstractYangian Y (sl (2))is realized in the bi-spin system coupled with a time-dependent external magnetic field.It is shown that Y (sl (2))generators can describe the transitions between the “spin triplet”and the “spin singlet”that evolve with time.Furthermore,new transition operators between the states with Berry phase factor and those between the states of Nuclear Magnetic Resonance (NMR)are presented.PACS:02.20.-a,03.65.-wKeywords:Yangian;Bi-spin system;Transition operator1.IntroductionYangian Algebras were established by Drinfeld[1,2]based on the investigation of Yang-Baxterequation.In recent years,many works in studying Yangian and its applications have been made, including the Yangian symmetry in quantum integrable models(such as Haldane-Shastry model [3],Calogero-Sutherland model[4],and Hubbard model[5,6])and the realization of Yangian in quantum mechanics[7,8,9].In quantum mechanics,the Yangian associated with sl(2)called Y(sl(2))has been realized in angular momentum quantum mechanics systems[7],hydrogen atom [8,9]and other systems.The sense of Yangian generators’transition operator has also been known. This is natural because Yangian algebras belong to hopf algebras and regard Lie algebras as their subalgebras.It was pointed out in[7]that,Y(sl(2))can be constructed in a bi-spin system with spinˆS1and ˆS2(ˆS k is the k th spin operator).It has been known that the Yangian generators{ˆI,ˆJ}can realize the transitions between the spin triplet and the spin singlet,which are the bases of quantum states of the bi-spin system(each S k has spin-1/2)that does not evolve with time.An interesting problem arises that if this system evolves with time,such as is coupled with a time-dependent external magneticfield,what will happen for the transition operators.A physical picture of the transition between the states is useful to help us understanding this time-dependent problem.In this paper,we will consider this issue.First,we will illustrate that Y(sl(2))generators{ˆI(t),ˆJ(t)}still play the role of transition operators.Then,we discuss the transitions between the states with Berryphase factor of this system under the adiabatic condition.Andfinally,using the combinationof the Y(sl(2))generators{ˆI(t),ˆJ(t)},the transition operators between the Nuclear Magnetic Resonance(NMR)states are also obtained.2.Realization of Y(sl(2))and transition operators for the bi-spin system in an external magneticfieldY(sl(2))is formed by a set of operators{ˆI,ˆJ}obeying the commutation relations[1,7]:[ˆIα,ˆIβ]=iǫαβγˆIγ,[ˆIα,ˆJβ]=iǫαβγˆJγ(α,β,γ=1,2,3),[ˆJ±,[ˆJ3,ˆJ±]]=ˆI±(ˆJ±ˆI3−ˆI±ˆJ3),[ˆJ3,[ˆJ+,ˆJ−]]=ˆI3(ˆI+ˆJ−−ˆJ+ˆI−).(1)ˆI stands for the generators of sl(2).Hereafter for any operatorsˆA±=ˆA1±iˆA2(ˆA=ˆI,ˆJ)areunderstood.In a bi-spin system with spinˆS1andˆS2,the Y(sl(2))generators take the form[7]:ˆI≡ˆS=ˆS1+ˆS2,ˆJ=µ(ˆS1−ˆS2)+ih√√2(µ−h4)|X00 ,ˆJ −|X 11 =−√4)|X 00 ,ˆJ +|X 00 =−√4)|X 11 ,ˆJ3|X 00 =(µ+h 2(µ+h2ˆS1·ˆS 2−γB (t )·ˆS (6)where γis gyromagnetic ratio.To study this system,we firstly write a simple hamiltonian de-scribing the bi-spin system in a steady magnetic field as follows:ˆH=−1B 0+B 3(t )B 01/2G (t )·ˆSi (i =1,2)(8)whereG (t )=(B 1(t ),B 2(t ),B 3(t )+B 0),(9)there exists the following transformation relation:ˆH(t )=ˆU (t )ˆH ˆU −1(t )(10)when we choose g=−γB0.From Eq.(10),it is shown that the eigenvalues ofˆH(t)are the same as those ofˆH,and the eigenstate|X jm(t)>ofˆH(t)can be got from the transformation of the eigenstate|X jm>ofˆH, i.e.,|X jm(t)>=ˆU(t)|X jm>(jm=11,10,1−1,00)(11) where|X jm>is the spin triplet(Eqs.(3))or the spin singlet(Eq.(4)).Through the unitary transformationˆU(t),the spin realization of Y(sl(2))has the time-dependentgeneratorsˆI(t)=ˆU(t)ˆIˆU−1(t),ˆJ(t)=ˆU(t)ˆJˆU−1(t),(12)which can be verified to still satisfy the definition Eqs.(1)of Y(sl(2)).ˆJα(t)(α=±,3)are the transition operators between the“spin triplet”(S=1){|X11(t)>,|X10(t)>,|X1−1(t)>}and the“spin singlet”(S=0)|X00(t)>in a time-dependent magneticfield,and the transition relationstake the same form as Eqs.(5):ˆJ+(t)|X1−1(t)>=√)|X00(t)>,4ˆJ3(t)|X10(t)>=(µ−hh2(µ−h2(µ+)|X10(t)>,4ˆJ−(t)|X00(t)>=√)|X1−1(t)>.(13)4In fact,the generatorsˆI(t)andˆJ(t)vary withˆU(t)because the selection ofˆU(t)is not exclusive.The choice ofˆU(t)does not affect the action of generators of Yangian Y(sl(2)),sowe can choose the unitary operatorˆU(t)as simple as possible.Taking advantages of the above results,we will give two physical applications in the following.3.Transition operators between the states with Berry phase factorIt has been shown that Berry phase[10]play a fundamental and an important role in quantum mechanics in the past two decades.Berry phase can be verified by experiments[11],and it can not be neglected in many physics lessons.Recently,Berry phase of the bi-spin system coupled with an external magneticfield has been studied[12,13].Now we will give the transition operators between the states with Berry phase factor.Consider the bi-spin system in a rotating magneticfield described byB1(t)=B0sinθcosω0t,B2(t)=−B0sinθsinω0t,B3(t)=B0cosθ(14) where constant B0is the strength(as referred to before)of the magneticfield.Substituting Eqs.(14)into Eqs.(8)and Eq.(11),we can give the eigenstates of Hamiltonian Eq.(6)that is the “triplet”(S=1)|X11(t)>=1√2(1−cosθ)e−2iω0t|X1−1>,|X10(t)>=12sinθe iω0t|X11>−cosθ|X10>−12sinθe−iω0t|X1−1>,|X1−1(t)>=1√2(1+cosθ)|X1−1>,(15) and the“singlet”(S=0)|X00(t)>=−|X00>.(16) Under the adiabatic condition,the state with Berry phase factor has the form:|ψjm(t)>=exp{−iwhere E jm is the eigenvalue of Hamiltonian Eq.(6)and is given byE11=−18,E1−1=−18.(18)On the other hand,the Berry phaseγjm(t)readsγ11(t)=ω0(1−cosθ)t,γ1−1(t)=−ω0(1−cosθ)t,γ10(t)=γ00(t)=0.(19) By comparing Eqs.(13)with Eq.(17),we immediatelyfind thatˆJ+(t)|ψ1−1(t)>=√4)exp{−i4)exp{−i2(µ−h¯h(E11−E00)t}exp{iγ11(t)}|ψ00(t)>,ˆJ+(t)|ψ00(t)>=−√4)exp{−i4)exp{i2(µ+h¯h(E1−1−E00)t}exp{−iγ1−1(t)}|ψ1−1(t)>.(20)We have solved the transition problems between the states with Berry phase factor by applying ˆJα(t)(α=±,3).4.Transition operators between the states of NMRNMR is a very important experiment technique based on quantum mechanics[14,15].It has been made rapid progress since1945.Very recently,NMR is used to realize the Geometric Quantum Computation[16,17].The motivation for this section is tofind the transition operators between the NMR states.We choose the same magneticfield and the Hamiltonian as that in the former sections.By solving the Schr¨o dinger equation and utilizing the magnetic resonance condition(MNC)ω0=γB3(21) we can get the states of NMR by choosing different initial states.These states are the time-dependent combination of the eigenstates of the Hamiltonian Eq.(6).The“triplet”(S=1)has the forms:|φ11(t)>=a1(t)|X11(t)>+a2(t)|X10(t)>+a3(t)|X1−1(t)>,|φ10(t)>=b1(t)|X11(t)>+b2(t)|X10(t)>+b3(t)|X1−1(t)>,|φ1−1(t)>=c1(t)|X11(t)>+c2(t)|X10(t)>+c3(t)|X1−1(t)>.(22) In the process of calculating,the eigenvalues of deformed wave functions under MNC Eq.(21)have the exact values:E′11=−18,E′1−1=−18.(23)The time-dependent coefficients in Eqs.(22)area1(t)=1¯hE′10t}exp{iω0t}(cosθcosω0t+cosω0t cosω1t+i sinω0t+i cosθsinω0t cosω1t+i sinθsinω1t),a2(t)=12exp{−i2exp{−i√¯h E′10t}exp{iω0t}(i cosω0t sinω1t−cosθsinω0t sinω1t+sinθcosω1t),b2(t)=−exp{−i√¯h E′10t}exp{−iω0t}(i cosω0t sinω1t+cosθsinω0t sinω1t−sinθcosω1t),c1(t)=1¯hE′10t}exp{iω0t}(−cosθcosω0t+cosω0t cosω1t+−i sinω0t+i cosθsinω0t cosω1t+i sinθsinω1t),c2(t)=12exp{−i2exp{−i2a1(t)(µ−h2a2(t)ˆJ3(t)}|φ00(t)>=√4)exp{−iE′00t}|φ11(t)>,ˆJ3(t)|φ10(t)>=b2(t)(µ−h2b2(t)ˆJ3(t)}|φ00(t)>=√4)exp{−iE′00t}|φ10(t)>,ˆJ+(t)|φ1−1(t)>=√4)exp{iE′00t}|φ00(t)>,{c3(t)ˆJ−(t)−c1(t)ˆJ+(t)+√2(µ+h5.ConclusionsIn summary,we get a time-dependent realization of Y(sl(2))in the bi-spin system coupledwith a time-dependent external magneticfield.Although we can verify that Y(sl(2))does notdescribe the symmetry of the system which we study becauseˆJ(t)does not commute withˆH(t), we concentre on the transition function of Yangian.Wefind that the generators{ˆI(t),ˆJ(t)}of Y(sl(2))can describe a new picture of transition between two quantum states at any time.Forbriefness,we have neglected the part of transitions that are described by the Lie algebra operators. As far as we know,many interesting investigations have relations with the bi-spin model coupled with a time-dependent external magneticfield,such as Geometric Quantum Computation[16,17], entanglement[18].At last,we emphasis that Yangian algebras belong to hopf algebras and take Lie algebras as their subalgebras,so the Yangian operators can connect the physical states with different Lie-algebra weights.It’s reasonable to believe that the more interesting Yangian realization and the more useful physical applications should be found.AcknowledegementWe thank Dr.Hong-Biao Zhang and Dr.Hui Jing for valuable discussions.This work is supported by NSF of China.References[1]V.G.Drinfeld,Soviet Math.Dokl.32(1985)254.[2]V.G.Drinfeld,Soviet Math.Dokl.36(1985)212.[3]F.D.M.Haldane,Z.N.C.Ha,J.C.Talstra,D.Bernard and V.Pasquier,Phys.Rev.Lett.69(1992)2021.[4]M.Wadati,Phys.Rev.Lett.60(1988)635.[5]D.B.Uglov and V.E.Korepin,Phys.Lett.A90(1994)238.[6]F.G¨o hmann and V.Inozemtsev,Phys.Lett.A214(1996)161.[7]M.L.Ge,K.Xue and Y.M.Cho,Phys.Lett.A249(1998)358.[8]M.L.Ge,K.Xue and Y.M.Cho,Phys.Lett.A260(1999)484.[9]C.M Bai,M.L.Ge and K.Xue,J.Stat.Phys.102(2001)545.[10]M.V.Berry,Proc.R.Soc.Lond.A392(1984)45.[11]R.Y.Chiao and Y.S.Wu,Phys.Rev.Lett.57(1986)933.[12]L.G.Yang and F.L.Yan,Phys.Lett.A265(2000)326.[13]F.L.Yan and L.G.Yang,Commun.Theor.Phys.35(2001)527.[14]C.P.Slichter,Principles of Nuclear Magnetic Resonance,Springer-Verlag,(1978),2nd ed.[15]C.T.Claude,K.Bernard and L.Franch,Quantum Mechanics,France:Hermam,John Wileyand Sons.Inc.(1977).[16]J.A.Jones,V.Vedral,A.Ekert and G.Castagnoli,Nature,403(2000)869.[17]J.A.Jones,Progress in Nuclear Magnetic Resonance Spectroscopy38(2001)325-360[18]R.G.Unanyan,N.V.Vitanov and K.Bergmann,Phys.Rev.Lett.87(2001)137902.11。

化工进展Chemical Industry and Engineering Progress2023 年第 42 卷第 10 期共价有机框架（COFs ）在锂离子电池中的应用马文杰，姚卫棠（成都大学机械工程学院，四川 成都 610100）摘要：锂离子电池（LIBs ）具技术成熟、能量密度较高、使用寿命长等优点使其在储能领域研究应用广泛，但传统商业化LIBs 由于本身电极材料及电解质的限制，存在可逆比容量有限、功率密度不高、循环性能较差、生产材料成本较高、工作过程面临安全隐患等不足。

本文简述了由轻质元素构成的晶态有机多孔材料共价有机框架（covalent organic frameworks ，COFs ），其有序的大孔道、可预先设计的结构、较大比表面积、低密度、易功能化等优势，完全具备利用于LIBs 关键材料的潜力。

回顾了近年来研究者们设计的各种COFs 及其在LIBs 的应用，包括COFs 在LIBs 电极材料、隔膜、电解质中的应用，得出COFs 应用于LIBs 具有卓越的电化学性能，最后对其在LIBs 的研究方向给予预测，以期为储能和再生能源产业的发展提供一定参考。

关键词：共价有机框架；多孔材料；锂离子电池；性能；再生能源；安全中图分类号：TQ152 文献标志码：A 文章编号：1000-6613（2023）10-5339-14Application of covalent organic frameworks ( COFs ) inlithium-ion batteriesMA Wenjie ，YAO Weitang(School of Mechanical Engineering, Chengdu University, Chengdu 610100, Sichuan, China)Abstract: Lithium-ion batteries (LIBs) have been widely used in the field of energy storage due to their advantages of mature technology, high energy density and long service life. However, due to the limitation of electrode materials and electrolytes, traditional commercial LIBs have the disadvantages of limited reversible specific capacity, low power density, poor cycle performance, high cost of production materials and potential safety hazards in the working process. In this paper, covalent organic frameworks (COFs), a crystalline organic porous material composed of light elements, were briefly described. Its ordered large pores, pre-designed structure, large specific surface area, low density, easy functionalization and other advantages fully had the potential to be used as key materials for LIBs. Various COFs designed by researchers in recent years and their applications in LIBs were reviewed, including the application of COFs in LIBs electrode materials, separators and electrolytes. It was concluded that COFs had excellent electrochemical properties when applied to LIBs, Finally. the research direction of COFs in LIBs was predicted in order to provide some reference for the development of energy storage and renewable energy industries.Keywords: covalent organic framework; porous material; lithium-ion battery; performance; renewable energy; safety综述与专论DOI ：10.16085/j.issn.1000-6613.2022-2041收稿日期：2022-11-02；修改稿日期：2023-01-07。

ARTESYN INTELLIGENT MP SERIESUp to 1500 WattsTotal Power:Up to 1500 WInput Voltage:85 - 264 Vac 120 - 300 Vdc# of Outputs:Up to 21DATA SHEETAdvanced Energy's Artesyn iMP1 series is an AC input to DC output configurable power system consisting of a microprocessor-controlled PFC front end providing seven slots that accept intelligent DC-DC converter modules with single, dual or triple outputs ranging from 2 V to 60 V. Single output modules come in four power ranges that can be mixed and matched and connected in parallel or series to obtain thousands of output combinations customized to any application.SPECIAL FEATURES⏹ Full medical EN60601 approval ⏹ Intelligent I 2C control⏹ Voltage adjustment on all outputs(manual or I 2C)⏹ Configurable input and output OKsignals and indicators⏹ Configurable inhibit/enable ⏹ Configurable output UP/DOWNsequencing⏹ Configurable current limit(foldback or constant current)⏹ High power density (8.8 W/cu-in)⏹ Intelligent fan (speed control/faultstatus)⏹ Downloadable GUI from website ⏹ Customer provided air option ⏹ uP controlled PFC input with activeinrush protection⏹ I 2C monitor of voltage, current, and⏹ Optional extended hold-up module(SEMI F47 compliance)⏹ Increased power density to 50% overstandard MP⏹ External switching frequency syncinput⏹ Optional conformal coating ⏹ Industrial temp range (-40 ºC to70 ºC)⏹ No preload required⏹ Industrial shock/vibration (> 50G’s)SAFETY⏹ UL UL60950/UL2601**⏹ CSA CSA22.2 No. 234 Level 5⏹ TUV EN60950/EN60601-1**⏹ BABT Compliance toEN60950/ EN60601 BS7002⏹ CB Certificate and report ⏹ CEMark to LVDIMPIMP* Can be controlled via I C** Controlled via I C but requires load calibrationINTERNAL PART NUMBER REFERENCE TABLEIMPOUTPUT MODULE LINE-UP* Note: Contact Factory for extended range down to 6 VIMP* Note: Contact Factory for extended range down to 6 V.** Increments of current not shown can be achieved by paralleling modules (add currents of each module selected).*** Total output power on dual model must not exceed 144 W.**** For single output modules only.Green reference lines indicate physical module groupingsOrdering Notes: Array 1. T he cases and modules of both MP and iMP series can be interchangedto allow more flexibility. If intelligent modules are used withnon-intelligent cases, a numeric code “4” is placed at the end of themodule code (ex. 4LL0 becomes 4LL4).2. USB to I2C module order code 73-769-001IMPSingle210 W360 W750 W1500 W ( 10 - 60 V)144 W 36 WDual 1500 W with Bus BarAdapter Option (used with the 10 - 60 V module)1500 W ( 2.0 - 8.0 V)TripleiMP4 (AC input on opposite side)iMP 4 = 2.5” x 5” x 10” 5 available slots(63.5 x 127 x 254 mm)Input90 - 264 Vac 180 - 264 Vac 750 W max. 1158 W max.iMP8 and iMP1iMP 1 onlyS L O T 7S L O T 5S L O T 6S L O T 4S L O T 3S L O T 2S L O T 1iMP 8 = 2.5” x 7” x 10” 6 available slots(63.5 x 177.8 x 254 mm)iMP 1 = 2.5” x 8” x 11” 7 available slots(63.5 x 203.2 x 279.4 mm)Input85 - 264 Vac 180 - 264 Vac 1000 W max. 1200 W max.1200 W max.1500 W max.IMPMates withMolex 90142-0010 Housing 90119-2110 Pin Connector Kit Part No.:70-841-004Mates withLandwin 2050S1000 Housing 2053T011V PinorJST PHDR-10VS Housing JST SPHD-002T-P0.5 (28-24) JST SPHD-001T-P0.5 (26-22)Connector Kit Part No.:70-841-023Figure 3. Connector J210651IMPiMP ModulesDC-DC Converter Output ModulesMates withMolex 90142-0010 Housing 90119-2110 PinFigure 4. Connector J1615 Figure 2. Connector J1Dual 144 WattSingle 1500 Watt 10-60 VSingle 750 WattDC OUTPUT CONNECTORV1 +V1 -V2 -V3 -V3 +V2 + 26-60-5060 PinSingle 1500 Watt 2-8 VTriple 36 WattIMP SERIES (CONTINUED)Notes:1. Input: Barrier type. Three No. 6-32 B.H. screws (0.375” centers). Max torque: 6 in-lbs. (0.67 N-m).2. C ontrol connectors: (J1) 10 position housing, gold plated contacts. Mates with Molex 90142-0010 housing with 90119-2110 crimp contacts (Molex C - Grid III Series) Connector kit includes mating connector and 10 pins, Astec part #70-841-004. (J2) 10 position housing (Landwin 2051P1000T). Mates with housing 2050S1000 (Landwin) with 2053T011P (Landwin) pins or JST PHDR-IOVS Housing and JST SPHD-002T-PO.5 pins.3. Chassis material: aluminum with chemical film coating (conductive).4. All dimensions are in millimeters and inches, and are typical.5. Customer mounting -3 sides M4, bottom also includes 8-32 mounting holes. Max. penetration is 0.150” (3.8 mm). Max. torque: 5 in-lbs. (0.57 N-m).6. O utput module connections: All single O/P modules are M4 x 8 mm screws. Max. torque: 10 in-lbs. (1.13 N-m).Dual O/P module is M3 x 8 mm screws. Max. torque: 5 in-lbs. (0.57 N-m).IMPiMP1 (1200/1500 Watts Max)8-Inch Case Size: iMP1: 2.5” x 8” x 11” (63.5 mm x 203.2 mm x 279.4 mm)Weight: i MP1 Case: 5.0 lb. • 210 W Single: 0.6 lb. • 360 W Single: 1.0 lb.• 750 W Single: 1.6 lb. • 144 W Dual: 0.6 lb.19.05MARKED AS XNotes:1. Input: Barrier type. Three No. 6-32 B.H. screws (0.375” centers). Max torque: 6 in-lbs (0.67 N-m).2. C ontrol connectors: (J1) 10 position housing, gold plated contacts. Mates with Molex 90142-0010 housing with 90119-2110 crimp contacts (Molex C - Grid III Series). Connector kit includes mating connector and 10 pins, Astec part #70-841-004. (J2) 10 position housing (Landwin 2051P1000T). Mates with housing 2050S1000 (Landwin) with 2053T011P (Landwin) pins or JST PHDR-IOVS Housing and JST SPHD-002T-PO.5 pins.3. Chassis material: aluminum with chemical film coating (conductive).4. All dimensions are in millimeters and inches, and are typical.5. Customer mounting -3 sides M4, bottom also includes 8-32 mounting holes. Max. penetration is 0.150” (3.8 mm). Max. torque: 5 in-lbs. (0.57 N-m).6. O utput module connections: All single O/P modules are M4 x 8 mm screws. Max. torque: 10 in-lbs.(1.13 N-m). Dual O/P module is M3 x 8 mm screws. Max. torque: 5 in-lbs. (0.57 N-m).IMPThe RS485/CAN-to-I 2C uses 2 Input Protocols and 1 Output Protocol.The Input Protocols used are RS485 using Modbus (Command Index:0x01), and CAN using modified Modus (Command Index: 0x02).The Output Protocol use is: I 2C withSMBus support (Command Index: 0x80).Master/Client Device(s)Output ProtocolsAdapter ProtocolInput ProtocolsAdapter CommandRS485 using ModbusAdapter ControlsCAN using Modi ed ModbusI 2C with SMBusSupportiMP CAN RS485RS485/CAN - to - I 2CFor detailed info, download the Software RequirementsSpecification (SRS) from /power/power-supplies/cat/101/Configurable-Power-Supplies-iMP-Series。

Recent developments in garnet based solid state electrolytes for thin filmbatteriesShiang Teng,Jiajia Tan,Ashutosh Tiwari ⇑Nanostructured Materials Research Laboratory,Department of Materials Science and Engineering,University of Utah,United Statesa r t i c l e i n f o Article history:Available online 1November 2013Keywords:Li ion batterySolid state electrolyte GarnetsIonic conductivityPulsed laser annealinga b s t r a c tThis paper reviews the current status of,and new progress in,the field of solid state electrolytes (SSE)for lithium ion batteries.In addition to a review of current technologies,we are also presenting our novel results on pulsed laser processing of garnet based SSEs,specifically Li 7La 3Zr 2O 12(LLZO).LLZO powders with a tetragonal structure were prepared by a sol–gel technique,then a pulsed laser annealing process was employed to covert the powders to cubic LLZO without any loss of lithium.The tetragonal LLZO exhibited a Li ion conductivity of 1.8Â10À7S/cm,whereas the laser annealed cubic LLZO showed a Li ion conductivity of 1.0Â10À4S/cm at room temperature.A systematic study of the effect of pulsed laser annealing (PLA)on the crystal structure,morphology,composition,and ionic conductivity of LLZO was performed via X-ray diffraction (XRD),scanning electron microscopy (SEM),energy dispersive X-ray spectroscopy (EDS),X-ray photoelectron spectroscopy (XPS),and electrochemical impedance spectros-copy (EIS)measurements.These results demonstrate that PLA is a powerful processing technique for syn-thesizing the high ionic conductivity cubic phase of LLZO at relatively low temperatures,as compared to conventional methods.Ó2013Elsevier Ltd.All rights reserved.1.IntroductionLithium ion batteries (LIB)are important for a wide variety of applications,spanning from portable electronics and hybrid auto-mobiles to large-scale electrical power storage systems [1,2].These batteries offer a number of advantages over other families of bat-teries,such as high energy density,long cycle lifetime,no memory effect and a wide range of operating temperatures.Currently the majority of LIBs are using liquid electrolytes due to their very high ionic conductivities,in the range of 10À3S/cm.However,these li-quid electrolytes suffer from very serious drawbacks such as flam-mability,leakages,and the formation of dendrites in the electrodes.In this respect,solid state Li ion conductors have garnered inter-est as substitutes for the liquid electrolytes.Solid state Li ion elec-trolytes are expected to offer several advantages over the currently commercialized liquid electrolytes such as higher thermal stability,absence of leakage and pollution,and a large electrochemical sta-bility window.In addition,the high elastic modulus in ceramics makes them suitable for the rigid thin film micro-batteries.Current solid-state Li ion conductors can be divided into four groups –NAS-ICON type,perovskite type,LiPON type and garnet type.The fol-lowing sections contain a brief introduction of each category. 1.1.NASICON structured Li ion conductorsThe NASICON (Na super ionic conductor)type solid electrolytes are of increasing interest because of their potential to replace li-quid electrolytes in LIBs.The general formula of a NASICON-type electrolyte is LiM 2(PO 4)3,where M =Ti,Ge or Hf.NASICON consists of a covalent skeleton M 2ðPO 4ÞÀ3containing MO 6octahedra and PO 4tetrahedra,as shown in Fig.1[3].The Li sites sit in the interstitials between the MO 6octehedra and PO 4tetrahedra [4].The main fac-tors that limit the use of NASICON type electrolytes are grain boundary effects,which behave as the scattering sites for ion transportation,thereby reducing the bulk ionic conductivity to 10À5S/cm 2[5].It has been reported that in Li 1+x Ti 2Àx R x (PO 4)3,the lithium ion conductivity can be increased through the substitution of Ti 4+by other elements [6–10](see Table 1).The substitution of the more stable Al 3+for the less stable Ti 4+increases the M–O bond strength and decreases the Li–O bond strength which results in the higher ionic conductivity [6,11–17].Among the various doped NASICON electrolytes,an optimal ionic conductivity of 1.3Â10À3S/cm has been reported for the simultaneous doping of B and Al in LiTi 2(PO 4)3[18].Despite this,the highest reported Li ion conductiv-ity in a NASICON based micro-battery utilizing Li 1.3Al 0.3Ti 1.7(PO 4)3thin film electrolyte is 2.7Â10À6S/cm (see Fig.2)with only a 0.2%capacity loss per charging cycle [19–21].Even though substi-tution provides an enhancement to the ionic conductivity,1359-0286/$-see front matter Ó2013Elsevier Ltd.All rights reserved./10.1016/j.cossms.2013.10.002Corresponding author.Tel.:+18015851666.E-mail address:tiwari@ (A.Tiwari).still possess many drawbacks,including the not stable when in contact with lithium metal rapid Ti 4+reduction [22–24],which greatly hinders LIBs.3crystal structure where blue spheres represent represent TiO 6,and red tetrahedral represent PO 4[4].for trivalent ion doped Li 1+x Ti 2Àx R x (PO 4)3compound.Li ionic conductivity at room temperature (S/cm)6Â10À41Â10À42Â10À43.8Â10À46Â10À4Nyquist plot for Li 1.3Al 0.3Ti 1.7(PO 4)3thin film at room temperature.permission from [21].Crystal structure of perovskite (ABO 3)type oxides.4.Nyquist plots for LLTO thin film with various post annealing temperatures.Reprinted with permission from [32].frequency semi-circle caused by the bulk LiPON film and a low frequency line caused by the interfaces between the film and electrodes.LiPON is widely used as an amorphous thin film electrolyte for the lithium micro-batteries with the highest reported ionic conductivity of 3.1Â10À6S/cm [39].In order address some of the shortcomings of LiPON based electrolytes,many strategies have been explored to synthesize LiPON with im-proved ionic conductivity and proper crystallinity [40–42].How-ever,most of the synthesis strategies do not produce crystalline LiPON.Recently,Senevirathne et al.have reported successful syn-thesis of a new crystalline LiPON with similar conductivity using solid state method [43].An important advantage of LiPON over other solid state electrolytes is its good stability and excellent cyclability [39,44].While LiPON based LIBs do show improvement over other battery classes,mainly due to their compatibility with lithium,their overall ionic conductivities have been limited.1.4.Garnet structured Li ion conductorsRecently,oxides with garnet related structures have gained a lot attention as the potential solid state electrolyte for LIBs.The typ-ical structure of garnets is A 3B 2C 3O 12with CO 4tetrahedra and BO octahedra that are connected via edge sharing [45].Although this the general formula,it has been reported that increasing the number of lithium per formula unit to five,such as in Li 5La 3B’2O 0=Bi,Sb,Na,Ta),results in a three orders of magnitude increase ionic conductivity [46].In this formula,La can be substituted group II elements in order to possess more Li,in which case it has ionic conductivity of 4Â10À5S/cm [22].A further increase the Li content per formula unit to 7,in Li 7La 3B 002O 12(B 00=Zr,Hf,Sn),has likewise attracted attention recently.The highest ionic conductivity in garnet-related structures has been reported to Fig. 5.Nyquist plots for LiPON thin film measured at various temperatures.Reprinted with permission from [39].Crystal structure of (a)cubic LLZO,(b)tetrahedral LiO LiO 6site in cubic Li 7La 3Zr 2O 12.Crystal structure of (a)tetragonal LLZO,(b)tetrahedral LiO 4site LiO 6sites in tetragonal Li 7La 3Zr 2O 12.Reprinted with permission32S.Teng et al./Current Opinion in Solid State and Materials Science18(2014)29–38Table2Methods for synthesizing LLZO.Synthesis method Excess of Li added Processing temperature(K)Phase formed Ionic conductivity(S/cm)at room temperature Activation energy(eV)Solid state[47]N/A1503K for36h Cubic3Â10À40.3Solid state[52]10%1253K for5h Tetragonal 1.6Â10À60.54Solid state[45]10%1503K for36h Cubic 1.8Â10À4N/ASol gel[53]N/A1346K for5h Tetragonal 3.1Â10À70.67Sol gel[49]10%1453K for36hours Cubic 5.6Â10À40.80Sol gel[49]10%1273K for20h Tetragonal 4.4Â10À70.54Fig.8.XRD patterns of LLZO as a function of number of laser shots with annealingfluence of125mJ/cm2per shot.Fig.9.XRD patterns of LLZO as the function of number laser shots for:(a)(211)peaks at2h=16.7°,(b)(420)peaks at2h=30.5°.lithium is required to compensate for lithium losses during high temperature sintering.2.3.The effect of extrinsic dopingSeveral reports have suggested that introducing extrinsic impu-rities into LLZO during processing would increase the ionic conduc-tivity.One type of impurity introduced in LLZO is aluminum.Some improvements in the formation of the cubic LLZO.First,the Al atom acts as a sintering aid to lower the processing temperature[57]. Second,Al atoms substitute in both the tetrahedral and octahedral Li sites,which increases the Li vacancy concentration,facilitating Li transport[58].In order to further elucidate the phase stability and conductivity behaviors of LLZO,the effects of other dopant cations such as Ga have been investigated by several groups.The reported ionic con-ductivities for Ga doped LLZO are in the range of10À4S/cm[59,60].Table4FWHM data for XRD peaks at2h=16.7o.Number of laser shots01234515 FWHM(deg.)0.37±0.0050.36±0.0050.38±0.0050.38±0.0050.36±0.0050.36±0.0050.35±0.005A cross sectional SEM image of the LLZO sample annealed with15125mJ/cm2laser pulses(annealed surface at bottom).Fig.11.Elemental mapping of lanthanum,zirconium,and oxygen on the laser annealed LLZO pellet.S.Teng et al./Current Opinion in Solid State and Materials Science18(2014)29–3833is very different from conventional heating time in pulsed laser annealing is on onds as opposed to furnace annealing ing and cooling does not provide sufficient evaporate from the sample via diffusion.escape from the LLZO during laser fusion coefficient.Very high local bly fast heating and quenching rates are pulsed laser annealing;this makes lasers 4.Cubic LLZO via pulsed laser annealingA sol–gel synthesis method was used to der.For this,Li 2CO 3(99%),La 2O 3(99%),and were dissolved in nitric acid to prepare a molar ratio between Li,La,and Zr was acid (99.5%)and ethylene glycol (in the added to the precursor solution to serve as pH of the precursor solution was adjusted resulting precursor solution was then turned to a solidified white powder.The and calcined at 1000°C for 10h (at 1°furnace.After calcination,the powder was eter pellet using a cold iso-static press 350MPa for ser annealingFor laser annealing,a KrF COMPex Pro from Lambda Physik Inc.was used.The and a pulse repetition rate of 1Hz was a uniform illumination,a beam the laser and the sample.The energy 125and 700mJ/cm 2.The size of the laser beam was kept slightly larger than the sample size so that the entire sample area was irra-the Rietveld analysis technique.Microstructure of the laser annealed samples was examined on the cross section surfaces using scanning electron microscopy (SEM)(Hitachi S-3000N).The qualitative elemental analysis was performed using energy dispersive X-ray spectroscopy (EDS).The lithium content was examined by performing X-ray photoelectron spectroscopy (XPS)with a Kratos Axis Ultra spectrometer using a monochromatic source.Room temperature ionic conductivity measurements were performed on the pressed pellets using a Gamry reference 600™Instrument over the frequency range of 1to 106Hz.For this,was sputter coated on the top and bottom surfaces of the pellet serve as Li ion blocking electrodes.Fig.12.EDS spectrums of laser annealed LLZO at:(a)bulk and (b)laser annealed surface.13.XPS spectrums for Li 1s of the (a)laser annealed and (b)unannealed LLZO samples.34S.Teng et al./Current Opinion in Solid State and Materials Science 18(2014)29–385.Results and discussionThe X-ray diffraction(XRD)patterns for the various LLZO sam-tetragonal structure as shown in Fig.9[55].As can be seen in Fig.9(a),with increased number of laser shots,the(211)peak splitting disappears and peak intensity increases.Similarly,withFig.14.Nyquist plots of the AC impedance spectra for:(a)the tetragonal LLZO and(b)the laser annealed cubic LLZO.sectional SEM images of the one laser shot at:(a)125mJ/cm2,(b)500mJ/cm2(c)600mJ/cm2and(d)700mJ/cm2laserS.Teng et al./Current Opinion in Solid State and Materials Science18(2014)29–3835and very close to the values expected in stoichiometric material (see Table 5).In order to monitor lithium content in the material,further characterization was performed using XPS.XPS is a very powerful technique that can detect elements with an atomic number of 3(lithium)and higher [67].Fig.13shows the Li 1S XPS spectrums of the laser annealed as well as the unannealed samples.A comparative study of these spectra showed similar lithium content in both of the samples.This indicates that even though the laser-annealing turns tetragonal LLZO to cubic phase,the composition of the material does not change.The room temperature AC conductivity results for the tetrago-nal and the laser-annealed cubic LLZO are shown in the complex impedance Nyquist plots in Fig.14.The solid lines in Fig.14represent fitted data based on an equivalent circuit model of (R total Q total )Q el ,where R total is the total resistance,Q total is the total constant phase element,and Q el is the constant phase element of the electrode,respectively [53,64].Both of the impedance spectra show a semicircle in the high frequency range which gives the total ionic resistance.The total ionic conductivity of the tetragonal LLZO pellet (diameter ~1",thickness ~1/12")is 1.8Â10À7S/cm which is comparable to the reported total ionic conductivity for the pure tetragonal LLZO [49].The total ionic conductivity of the laser an-nealed cubic LLZO pellet (diameter ~1/2",thickness ~1/5")is 1.6Â10À5S/cm which is about two orders of magnitude higher than that for the tetragonal LLZO.However,still the total ionic con-ductivity of the laser annealed cubic LLZO is lower than the values reported by Shimonishi et al.and Murugan et al.[47,49].The low ionic conductivity for the laser annealed cubic LLZO was attributed to the fine grain structure of the material which results in a low density of 4.55g/cm 3compared to the theoretical value of 5.098g/cm 3[52].In general,the solid state ionic conductivity has a close relation-ship with grain size;the material with large grains and fewer grain boundaries usually has higher conductivity.To improve the ionic conductivity of the laser annealed LLZO,a follow up experiment was conducted with higher laser pulse energies.It was observed that when laser energy is increased,grain starts to grow (see Fig.15).The average grain size was found to increase from 4l m for 125mJ/cm 2to 8l m for 600mJ/cm 2.On further increasing the laser pulse energy to 700mJ/cm 2,slight melting at the grain boundaries occurred which enhanced the connectivity between the grains.Fig.16shows EIS data for the sample that had been an-nealed by 3laser shots at 600mJ/cm 2.The total ionic conductivity of this sample (diameter ~1/2",thickness ~1/7")was measured as1.0Â10À4S/cm,which is comparable to the highest ionic conduc-tivity for the pure cubic LLZO reported by Murugan et al.[47].6.ConclusionRechargeable lithium-ion batteries have emerged as the most prevalent medium for electricity storage in portable electronic de-vices due to their light weight,flexibility,and high energy density.These batteries are also considered very useful for their potential application in next-generation hybrid electric vehicles.However,a major drawback that plagues the application of these batteries is their flammability,specifically that of liquid electrolytes.To overcome the aforementioned shortcoming,an extensive amount of research has been performed to develop high ionic-conductivity solid state electrolytes.A wide spectrum of electrolyte materials reviewed here demonstrates that LLZO garnets are the most favor-able electrolyte materials for designing solid-state lithium-ion bat-teries,both due to their high ionic conductivities and overall compatibility with current LIBs,specifically Li metal.LLZO exists in two crystallographic phases,cubic as well as tetragonal.Cubic phase of LLZO exhibits higher ionic conductivity compared to tetragonal phase.However,the synthesis of cubic LLZO requires sintering at very high temperatures which results in lithium deficiencies and also limits its application.A way to overcome this problem is by doping LLZO with extrinsic dopants such as Al or Ga which lowers the sintering temperature required to achieve cubic phase.However,the reduction in sintering temperature comes with a new problem of secondary phase precipitation which can hinder the ion transport in the material.In the present study,we have developed a pulsed laser annealing process of converting tetragonal LLZO to cubic LLZO at room temperature without any extrinsic dopants.The room temperature ionic conductivity for the laser annealed cubic LLZO yields a three orders of magnitude improve-ment over the tetragonal LLZO.Selected references⁄⁄of very special interest:[⁄⁄18]Found a high ionic conductivity of 1.3Â10À3S/cm by simultaneous doping with B and Al in Li 1.3Al 0.3Ti 1.7(PO 4)3.[⁄⁄26]Found adding a small amount of Li 7La 3Zr 2O 12sol in the precursor increases ionic conductivity of Li 0.35La 0.55TiO 3.[⁄⁄43]A novel new crystalline LiPON was investigated.[⁄⁄50]A comprehensive understanding of the phase stability for Li 7La 3Zr 2O 12.[⁄⁄55,⁄56]A comprehensive understanding of the role of Al in Al doped Li 7La 3Zr 2O 12.⁄of special interest:[⁄16]Study spark plasma sintering method to reach 97%theo-retical density and high ionic conductivity for LiTi 2(PO 4)3.[⁄17]An investigation on various Al sources on Li ion conductiv-ity for Li 1.3Al 0.3Ti 1.7(PO 4)3[⁄19]A comparison of ionic conductivity of Li 1.3Al 0.3Ti 1.7(PO 4)3film prepared by different annealing method (rapid thermal vs.conventional furnace annealing).[⁄24]A comparison of ionic conductivity of Li 7La 3Zr 2O 12sin-tered in different ambient (in air or Ar flow).[⁄27,⁄28,⁄30]Found the ionic conductivity of perovskite oxides is very sensitive to the lithium content and A-site vacancies.[⁄32]Found post annealing Li 0.35La 0.55TiO 3film increases the ionic conductivity.[⁄38]Investigation of ionic conductivity of LiPON thin film pre-pared under a N 2–O 2–Ar plasmaenvironment.16.Nyquist plots of the AC impedance spectrum for the laser annealed cubic LLZO with large grain.and Materials Science 18(2014)29–38[⁄40]A comprehensive understanding of the mechanism of nitrogen in ionic conductivity for LiPON thinfilm.[⁄48]Direct observation of impurity phases at the grain bound-ary and their effects on ionic conductivity for Al doped Li7La3Zr2O12.[⁄59,⁄60]Investigation of Li ion conductivity of Ga doped Li7La3Zr2O12.AcknowledgementFinancial Support from the NASA EPSCoR of Utah and Petroleum Research Fund(PRF)is thankfully acknowledged.References[1]Goodenough JB,Park K-S.The Li-ion rechargeable battery:a perspective.J AmChem Soc2013;135:1167–76.[2]de las Casas C,Li W.A review of application of carbon nanotubes for lithiumion battery anode material.J Power Sources2012;208:74–85.[3]Knauth P.Inorganic solid Li ion conductors:an overview.Solid State Ionics2009;180:911–6.[4]Key B,Schroeder DJ,Ingram BJ,Vaughey JT.Solution-based synthesis andcharacterization of lithium-ion conducting phosphate ceramics for lithium metal batteries.Chem Mater2011;24:287–93.[5]Gellert M,Gries KI,Yada C,Rosciano F,Volz K,Roling B.Grain boundaries in alithium aluminum titanium phosphate-type fast lithium ion conducting glass ceramic:microstructure and nonlinear ion transport properties.J Phys Chem C 2012;116:22675–8.[6]Nagata K,Nanno T.All solid battery with phosphate compounds made throughsintering process.J Power Sources2007;174:832–7.[7]Peng H,Xie H,Goodenough e of B2O3to improve Li+-ion transport inLiTi2(PO4)3-based ceramics.J Power Sources2012;197:310–3.[8]Bounar N,Benabbas A,Bouremmad F,Ropa P,Carru J-C.Structure,microstructure and ionic conductivity of the solid solution LiTi2Àx Sn x(PO4)3.Physica B2012;407:403–7.[9]Johnson P,Sammes N,Imanishi N,Takeda Y,Yamamoto O.Effect ofmicrostructure on the conductivity of a NASICON-type lithium ion conductor.Solid State Ionics2011;192:326–9.[10]Jadhav HS,Cho M-S,Kalubarme RS,Lee J-S,Jung K-N,Shin K-H,et al.Influencesof B2O3addition on the ionic conductivity of Li1.5Al0.5Ge1.5(PO4)3glass ceramics.J Power Sources2013.[11]Arbi K,Lazarraga MG,Ben Hassen Chehimi D,Ayadi-Trabelsi M,Rojo JM,Sanz J,et al.Lithium mobility in Li1.2Ti1.8R0.2(PO4)3compounds(R=Al,Ga,Sc,In)as followed by NMR and impedance spectroscopy.Chem Mater2003;16:255–62.[12]Arbi K,Mandal S,Rojo JM,Sanz J.Dependence of ionic conductivity oncomposition of fast ionic conductors Li1+x Ti2Àx Al x(PO4)3,06x60.7.A parallel NMR and electric impedance study.Chem Mater2002;14:1091–7.[13]Huang L,Wen Z,Wu M,Wu X,Liu Y,Wang X.Electrochemical properties ofLi1.4Al0.4Ti1.6(PO4)3synthesized by a co-precipitation method.J Power Sources 2011;196:6943–6.[14]Takada K,Tansho M,Yanase I,Inada T,Kajiyama A,Kouguchi M,et al.Lithiumion conduction in LiTi2(PO4)3.Solid State Ionics2001;139:241–7.[15]Duluard S,Paillassa A,Puech L,Vinatier P,Turq V,Rozier P,et al.Lithiumconducting solid electrolyte Li1.3Al0.3Ti1.7(PO4)3obtained via solution chemistry.J Eur Ceram Soc2013;33:1145–53.[16]Kotobuki M,Koishi M.Preparation of Li1.5Al0.5Ti1.5(PO4)3solid electrolyte via asol–gel route using various Al sources.Ceram Int2013;39:4645–9.[17]Xiao Z-b,Chen S,Guo M-m.Influence of Li3PO4addition on properties oflithium ion-conductive electrolyte Li1.3Al0.3Ti1.7(PO4)3.T Nonferr Metal Soc 2011;21:2454–8.[18]Chen H,Tao H,Wu Q,Zhao X.Crystallization kinetics of superionic conductiveAl(B,La)-incorporated LiTi2(PO4)3glass-ceramics.J Am Ceram Soc 2013;96:801–5.[19]Wu X,Chen S,Mai F,Zhao J,He Z.Influence of the annealing technique on theproperties of Li ion-conductive Li1.3Al0.3Ti1.7(PO4)3films.Ionics 2013;19:589–93.[20]Wu X,Liu J,Li R,Chen S,Ma M.Preparation and characterization of LiMn2O4/Li1.3Al0.3Ti1.7(PO4)3/LiMn2O4thin-film battery by spray technique.Russ J Electrochem2011;47:917–22.[21]Popovici D,Nagai H,Fujishima S,Akedo J.Preparation of lithium aluminumtitanium phosphate electrolytes thickfilms by aerosol deposition method.J Am Ceram Soc2011;94:3847–50.[22]Ohta S,Kobayashi T,Asaoka T.High lithium ionic conductivity in the garnet-type oxide Li7ÀX La3(Zr2ÀX,Nb X)O12.J Power Sources2011;196:3342–5. [23]Kotobuki M,Kanamura K,Sato Y,Yoshida T.Fabrication of all-solid-statelithium battery with lithium metal anode using Al2O3-added Li7La3Zr2O12solid electrolyte.J Power Sources2011;196:7750–4.[24]Kotobuki M,Kanamura K,Sato Y,Yamamoto K,Yoshida T.Electrochemicalproperties of Li7La3Zr2O12solid electrolyte prepared in argon atmosphere.J Power Sources2012;199:346–9.[25]Emery J,Buzare JY,Bohnke O,Fourquet JL.Lithium-7NMR and ionicconductivity studies of lanthanum lithium titanate electrolytes.Solid State Ionics1997;99:41–51.[26]Chen K,Huang M,Shen Y,Lin Y,Nan CW.Improving ionic conductivity ofLi0.35La0.55TiO3ceramics by introducing Li7La3Zr2O12sol into the precursor powder.Solid State Ionics2013;235:8–13.[27]Suthirakun S,Ammal SC,Xiao G,Chen F,Huang K,zur Loye H-C,et al.Obtaining mixed ionic/electronic conductivity in perovskite oxides in a reducing environment:a computational prediction for doped SrTiO3.Solid State Ionics2012;228:37–45.[28]Bucheli W,Durán T,Jimenez R,Sanz J,Varez A.On the influence of the vacancydistribution on the structure and ionic conductivity of A-site-deficient Li x Sr x La2/3Àx TiO3perovskites.Inorg Chem2012;51:5831–8.[29]Li X,Zhao H,Luo D,Huang K.Electrical conductivity and stability of A-sitedeficient(La,Sc)co-doped SrTiO3mixed ionic-electronic conductor.Mater Lett 2011;65:2624–7.[30]Mather GC,García-Martín S,Benne D,Ritter C,Amador U.A-site-cationdeficiency in the SrCe0.9Yb01O3Àd perovskite:effects of charge-compensation mechanism on structure and proton conductivity.J Mater Chem 2011;21:5764–73.[31]Šalkus T,Kazakevicˇius E,Kezˇionis A,Orliukas AF,Badot JC,Bohnke O.Determination of the non-Arrhenius behaviour of the bulk conductivity of fast ionic conductors LLTO at high temperature.Solid State Ionics 2011;188:69–72.[32]Xiong Y,Tao H,Zhao J,Cheng H,Zhao X.Effects of annealing temperature onstructure and opt-electric properties of ion-conducting LLTO thinfilms prepared by RF magnetron sputtering.J Alloys Compd2011;509:1910–4. [33]Kumatani A,Ohsawa T,Shimizu R,Takagi Y,Shiraki S,Hitosugi T.Growthprocesses of lithium titanate thinfilms deposited by using pulsed laser deposition.Appl Phys Lett2012;101:123103–4.[34]Yuan T,Cai R,Wang K,Ran R,Liu S,Shao bustion synthesis of high-performance Li4Ti5O12for secondary Li-ion battery.Ceram Int 2009;35:1757–68.[35]Bates JB,Dudney NJ,Neudecker B,Ueda A,Evans CD.Thin-film lithium andlithium-ion batteries.Solid State Ionics2000;135:33–45.[36]Nimisha CS,Rao KY,Venkatesh G,Rao GM,Munichandraiah N.Sputterdeposited LiPON thinfilms from powder target as electrolyte for thinfilm battery applications.Thin Solid Films2011;519:3401–6.[37]Xu F,Dudney NJ,Veith GM,Kim Y,Erdonmez C,Lai W,et al.Properties oflithium phosphorus oxynitride(LiPON)for3D solid-state lithium batteries.J Mater Res2010;25:1507–15.[38]Meda L,Maxie EE.Lipon thinfilms grown by plasma-enhanced metalorganicchemical vapor deposition in a N2–H2–Ar gas mixture.Thin Solid Films 2012;520:1799–803.[39]Suzuki N,Shirai S,Takahashi N,Inaba T,Shiga T.A lithium phosphorousoxynitride(LiPON)film sputtered from unsintered Li3PO4powder target.Solid State Ionics2011;191:49–54.[40]Mascaraque N,Fierro JLG,Durán A,Muñoz F.An interpretation for the increaseof ionic conductivity by nitrogen incorporation in LiPON oxynitride glasses.Solid State Ionics2013;233:73–9.[41]Kim YG,Wadley HNG.The influence of the nitrogen-ionflux on structure andionic conductivity of vapor deposited lithium phosphorus oxynitridefilms.J Power Sources2011;196:1371–7.[42]Fleutot B,Pecquenard B,Martinez H,Letellier M,Levasseur A.Investigation ofthe local structure of LiPON thinfilms to better understand the role of nitrogen on their performance.Solid State Ionics2011;186:29–36.[43]Senevirathne K,Day CS,Gross MD,Lachgar A,Holzwarth NAW.A newcrystalline LiPON electrolyte:Synthesis,properties,and electronic structure.Solid State Ionics2013;233:95–101.[44]Ribeiro JF,Sousa R,Carmo JP,Gonçalves LM,Silva MF,Silva MM,et al.Enhanced solid-state electrolytes made of lithium phosphorous oxynitride films.Thin Solid Films2012;522:85–9.[45]Masashi K,Hirokazu M,Kiyoshi K,Yosuke S,Toshihiro patibility ofLi7La3Zr2O12solid electrolyte to all-solid-state battery using Li metal anode.J Electrochem Soc2010;157:A1076–9.[46]Murugan R,Weppner W,Schmid-Beurmann P,Thangadurai V.Structure andlithium ion conductivity of bismuth containing lithium garnets Li5La3Bi2O12 and Li6SrLa2Bi2O12.Mater Sci Eng B2007;143:14–20.[47]Murugan R,Thangadurai V,Weppner W.Fast lithium ion conduction ingarnet-type Li7La3Zr2O12.Angew Chem Int Ed2007;46:7778–81.[48]Jin Y,McGinn PJ.Al-doped Li7La3Zr2O12synthesized by a polymerized complexmethod.J Power Sources2011;196:8683–7.[49]Shimonishi Y,Toda A,Zhang T,Hirano A,Imanishi N,Yamamoto O,et al.Synthesis of garnet-type Li7Àx La3Zr2O12Àx and its stability in aqueous solutions.Solid State Ionics2011;183:48–53.[50]Bernstein N,Johannes MD,Hoang K.Origin of the structural phase transition inLi7La3Zr2O12.Phys Rev Lett2012;109:205702.[51]Xie H,Alonso JA,Li Y,Fernández-Díaz MT,Goodenough JB.Lithiumdistribution in aluminum-free cubic Li7La3Zr2O12.Chem Mater 2011;23:3587–9.[52]Awaka J,Kijima N,Hayakawa H,Akimoto J.Synthesis and structure analysis oftetragonal Li7La3Zr2O12with the garnet-related type structure.J Solid State Chem2009;182:2046–52.[53]Kokal I,Somer M,Notten PHL,Hintzen HT.Sol–gel synthesis and lithium ionconductivity of Li7La3Zr2O12with garnet-related type structure.Solid State Ionics2011;185:42–6.S.Teng et al./Current Opinion in Solid State and Materials Science18(2014)29–3837。

Layer-Based Dependency Parsing*Ping Jian and Chengqing ZongInstitute of Automation, Chinese Academy of SciencesNo. 95 Zhongguancun East Road, Beijing, 100190, China{pjian, cqzong}@Abstract. In this paper, a layer-based projective dependency parsing approach is presented.This novel approach works layer by layer from the bottom up. Inside the layer thedependency graphs are searched exhaustively while between the layers the parser statetransfers deterministically. Taking the dependency layer as the parsing unit, the proposedparser has a lower computational complexity than graph-based models which search for awhole dependency graph and alleviates the error propagation that transition-based modelssuffer from to some extent. Furthermore, our parser adopts the sequence labeling models tofind the optimal sub-graph of the layer which demonstrates that the sequence labelingtechniques are also competent for hierarchical structure analysis tasks. Experimental resultsindicate that the proposed approach offers desirable accuracies and especially a fast parsingspeed.Keywords: dependency parsing, dependency layer, sequence labeling1IntroductionGraph-based models (McDonald et al., 2005; McDonald and Pereira, 2006) and transition-based models (Yamada and Matsumoto, 2003; Nivre and Scholz, 2004) are two dominant paradigms in the dependency parsing community. McDonald and Nivre (2007) have made elaborate analyses about the very different theoretical properties of these two kinds of models and the corresponding experimental behaviors.Generally, graph-based approaches learn a model for scoring possible dependency graphs of an input sentence and apply exhaustive search algorithms to find the one that maximizes the score. The unit graph-based models calculate is the whole sentence (the whole dependency graph) both in training and inference procedures, which results in a cubic computational complexity (in projective case). By contrast, transition-based approaches train a classifier to greedily choose the best parsing action under the current parser state. They make decisions at a configuration which is usually composed by a couple of focus tokens and the parsing contexts. Therefore, these two kinds of dependency parsing methods represent the two extremes when they seek the best dependency structure of the input sentence. In this paper, we adopt a moderate structural granularity to calculate the parser: a dependency layer.The dependency layer we mean here is a set of tokens whose dependency depth (the depth of the dependency tree) is at most one. Inside the layer the dependency graphs can be searched exhaustively while between the layers the parser state transfers deterministically. On one hand, this design will decrease the computational cost for searching the whole tree like graph-based models do; on the other hand, it may alleviate the error propagation resulting from the complete no “search” outside the parsing configuration in transition-based models.*The research work has been partially funded by the Natural Science Foundation of China under grant No.60736014, 60723005 and 90820303, the National Key Technology R&D Program under grant No.2006BAH03B02, the Hi-Tech Research and Development Program (863 Program) of China under grant No.2006AA010108-4, and also supported by the China-Singapore Institute of Digital Media as well.Copyright 2009 by Ping Jian and Chengqing Zong23rd Pacific Asia Conference on Language,Information and Computation,pages230–239It is well known that chunking, which is deemed to be a useful and tractable precursor to full parsing, has been successfully handled by sequence labeling techniques (Kudo and Matsumoto, 2001; Sha and Pereira, 2003). Inspired by this scheme, we adopt the globally optimal sequence labeling to search the best depth-one sub-graph in the dependency layer. We believe that the line-typed sequential models are potent complementarities to the tree-typed hierarchical ones or even the latent substitutes.The experiments show that our layer-based parser yields comparable dependency attachment accuracies to the state-of-the-art dependency parsers on both English and Chinese datasets. Especially, it is quite efficient due to the layer-based search and sequence typed analysis. The remainder of the paper is organized as follows: Section 2 describes the details of the algorithm and feature set. Section 3 presents the experimental results. The related work is discussed in Section 4. Conclusion and future work comprise Section 5.2 Layer-based Parsing Approach2.1 AlgorithmsWu et al. (2007) designed a neighbor parser to identify the neighboring parent-child relations between two consecutive tokens in the input sentence. Following their framework we label the dependency relations in our parsing layer. An example is shown in Figure 1(a). The first and second columns represent the words and part-of-speech (POS) tags respectively. The third column implies whether the token modifies its left neighbor (LH, left-headed) or right neighbor (RH, right-headed) or neither (O). The string behind the character “_” indicates the dependency type of the neighboring link.Wu et al. (2007) employed linear chain conditional random fields (CRFs) as the labelingalgorithm to capture the higher order features and avoid the greedy search when labeling with sequential classifiers (Cheng et al., 2006). To prevent the error propagation, they regarded the labeling results as features of the subsequent parsing stage instead of reducing the child words. However, this weakens the strength that neighboring parsing can provide. In our approach, besides the CRF-based relation labeler, an additional tagger is introduced to examine whether a dependent child can be reduced, i.e., whether it has found its head and has already been a complete sub-tree. The reduce tagger tries to guarantee safe reductions and ensures the parsed structures can be formed into a tree after several passes of analysis. In Figure 1(b), the letter “r” in the rightmost column implies that the corresponding token will be reduced while others are reserved for the next stage.The reduce tagger is also trained by linear chain CRFs to fulfill the globally optimal property of the layer-based labeling. Specially, when continuous attachments happen in the same direction, only the lowest child token is reduced although other tokens in this chain are complete sub-trees after the current labeling. This enables the tokens to change their Figure 1: Example of (a) the sequential neighboring relation labeling, (b) the reduce decision labeling.TheDT RH_NMOD chairNN O ofIN LH_NMOD theDT RH_NMOD conferenceNN O declaredVBD O thatDT RH_NOMD decision NN OThe chair of the conference declared that decision(b)The DT RH_NMOD r chair NN O o of IN LH_NMOD o the DT RH_NMOD r conference NN O o declared VBD O o that DT RH_NOMD r decision NN O o (a)attachments when the context is refreshed in the next layer. For example, in Figure 2, if the parent word “of” and the child word “conference” are far from each other in the early parsing stage, the child “conference” may be wrongly attached to the word “declared” (Figure 2(a)) because long distance interrelations are difficult to be caught in sequence labeling models. If the tagger is learned to only reduce the lowest child token each time, i.e., the leftmost word “the”, the word “conference” has the chance to adjoin “of” and be attached correctly at last (Figure 2(b)).As the dependency relations only exist between adjacent tokens and all the survivals will berelabeled in the next layer, the dependency depth of the layer is at most one.Pseudo-code of the parsing algorithm is given as follows:Input sentence: w 1, w 2, …, w n Initialize: L = {w 1, w 2, …, w n };have_reduce = false ; Start: While |L|>1 do begin x = get_feature (L);y 1 = estimate_relation (model 1, x);y 2 = estimate_reduce (model 2, x, y 1);have_reduce = sign(count_reduce (y 2));if(have_reduce == true )reduce (L, y 2);have_reduce = false ; else break; end; end.At each processing stage, two functions, estimate_relation and estimate_reduce , are employed to label the sequence L with neighboring dependency relations (y 1) and reduce decisions (y 2). model 1 and model 2 are the pre-trained models accordingly. Then the parser reduces the “r” tagged tokens and transfers them as the children features for the next labeling stage. This process is repeated until there is no token to be reduced or the size of L equals 1. The remaining parsing process for the example sentence in Figure 1 is illustrated step by step in Figure 3 to give a more specific description of the algorithm.Together with the initial labeling stage showed in Figure 1(b), the layer-based algorithm spends five iterations, i.e., five layers to get the final dependency graph of the input sentence. In each layer, the neighboring dependency relations and reduce decisions are traded off at different chair the - Oo of - - LH_NMODo conference the - LH_PMODr declared - - Oo decision that - LH_OBJ r chair the - O o of - conference LH_NMOD r declared - decision O o chair the of RH_SUB rdeclared - decision O odeclared chair decision O o Figure 3: The parsing process following th e stage showed in Figure 1(b). The second column lists the left child of the current token attached in the latest analysis and the third column is the right one. of… theRH_NMOD conferenceRH_SUB × declared … of … conference LH_PMOD √ declared …(a): (b): Figure 2: Long dependency attaching error in neighboring relation labelingsequence positions to obtain a globally optimal depth-one dependency sub-graph. Between the layers, the pre-built structure is handed on through the surviving tokens as well as their children. Since dependency relations only exist between two consecutive tokens, the child appearing in the observation sequence is always the leftmost or rightmost one of the parent token. Previous work based on deterministic models (Nivre and Scholz, 2004; Hall et al., 2007) has verified that the information of the children at these positions is more useful than that of others.For training, the parsing process described above is repeated on each sentence in the training set to pick up instances on different layers.In addition, the reduce examiner in the two-time labeling algorithm described above relies too much on the relation labeling results since it takes the relation labels as features. Therefore, a one-time labeling framework is introduced to be an alteration of the two-time labeling one. Figure 4 shows an example. The strings in the third column are the integrated symbols of the dependency relation labels and the reduce labels. Because the token whose head is not found will not be tagged with “r”, a unique symbol “O” is enough to express this case.The DT RH_NMOD_rchair NN Oof IN LH_NMOD_othe DT RH_NMOD_rconference NN Odeclared VBD Othat DT RH_NOMD_rdecision NN OFigure 4: Integration of the relation and reduce labels2.2Usage of N-best Searching ResultsThe algorithm described above stops the parsing process if there is no reduce label “r” in the current layer. However, sometimes the fact is that the parser quits so early while the tree is not well formed yet at that point. One reason is that the reduce tagger is more prone to assign an “o” than an “r” due to the unbalanced training instances. Taking this into account, we use the n-best searching results produced by the CRF-based labeler to amend.Taking the two-time labeling for example, although there is no “r” assigned in the current stage, the parsing process still continues if there is a relation annotated between the neighbors. The parser will ask for the next best relation label sequence (y1’) and consequently estimate the reduce labels based on it. But if y1’ is not assigned with relations, the parser will fall back on the initial best labels (y1) and further request the next best reduce labels for y1. In our experiments, only 2-best outputs of the labeler are utilized and the experimental results show that it works well.2.3Feature DesignThe features used in our labelers are summarized in Table 1. Features of the tokens and children are prepared to parameterize the dependency attachment model. The relation features are added when tagging the reduce decisions in two-time labeling case.As a typical sequence labeling task, the features chosen for our parser are similar to those adopted in (Sha and Pereira, 2003) for shallow parsing, and a first-order Markov dependency between labels is considered.Cheng et al. (2006) argued that the features and the strategies for parsing in the early stage are different from parsing in the upper stages in bottom-up deterministic parsing approaches. Because the initial stage parses “words” while the upper stages parse “phrases”. For this reason, we improve the proposed parser to a model-divided one in which one model is only for the first parsing layer and the other takes charge of the higher layers. The children features listed in Table 1 will not be used to parameterize the first layer model.Table 1: Feature set for the neighboring parsing. w is the word and p is the POS tag of the token. lc and rc represent the leftmost and rightmost child, and the dependency relation type of them uses typ. The relation features like “RH_SUB” are denoted by rel. Digit bracketed marks the position of the token where the feature is sampled, negative for the left and positive for the right. “ ” denotes the combination.Tokens w[-3], w[-2], w[-1], w[0], w[1], w[2], w[3]p[-3], p[-2], p[-1], p[0], p[1], p[2], p[3]p[-2] p[-1], p[-1] p[0], p[0] p[1], p[1] p[2], p[-1] p[0] p[1]w[-1] p[-1], w[0] p[0], w[1] p[1]Children w_lc[0], w_rc[0]p_lc[-1], p_rc[-1], p_lc[0], p_rc[0], p_lc[1], p_rc[1]p[-1] p_lc[-1], p[-1] p_rc[-1], p[0] p_lc[0], p[0] p_rc[0], p[1] p_lc[1], p[1] p_rc[1]typ_lc[-1], typ_rc[-1], typ_lc[0], typ_rc[0], typ_lc[1], typ_rc[1]Relations rel[-3], rel[-2], rel[-1], rel[0], rel[1], rel[2], rel[3]3ExperimentsTo evaluate the effectiveness and efficiency of the layer-based approach, we conducted dependency parsing experiments on both English and Chinese datasets.The English experiments were carried out on the WSJ part of Penn Treebank (Marcus et al., 1993). To match the previous work (Nivre and Scholz, 2004; Hall et al., 2006; McDonald and Pereira, 2006), we used sections 02-21 for training, section 22 for development and section 23 (about 56,684 words) for testing. The head-finding rules employed by Yamada and Matsumoto (2003) were adopted here to convert the constituent structures to dependency ones and a set of 12 dependency types was utilized as what Hall et al. (2006) did.1 The POS tags for the development and testing set were automatically assigned by MXPOST (Ratnaparkhi, 1996). A tagging accuracy 97.05% was achieved on the testing set.The Chinese experiments were evaluated on the Penn Chinese Treebank (CTB) version 5.0 (Xue et al., 2005). The corpus was split into training, development, and testing data as Duan et al. (2007) did to balance the different resources. 16,079 sentences were for training, 803 for development, and 1,905 (about 50,319 words) for testing. The head-finding rules and dependency type set also followed Hall et al. (2006). 2 Gold standard POS tags were used.Eight parsers involved in our main experiments are concisely introduced as following: MaltParser (Nivre et al., 2006): adopts transition-based model described in (Nivre, 2004). Here, MaltParser version 1.1 is employed.Yamada03: our implementation of another typical transition-based model proposed in (Yamada and Matsumoto, 2003).MSTParser1: The first-order paradigm of MSTParser3 which implements the graph-based models described in (McDonald et al., 2005; McDonald and Pereira, 2006). Version 0.2 is used. MSTParser2: The second-order paradigm of MSTParser.Duan07: A probabilistic parsing action model proposed by Duan et al. (2007) which globally seeks the optimal action sequence above the transition-based model described in (Yamada and Matsumoto, 2003) with beam search algorithm and employs SVMs for learning.LDParser1: One of the layer-based dependency parsers which labels the relations and reduce decisions at one time.LDParser2: One of the layer-based dependency parsers which labels the relations and reduce decisions separately.1The tree conversion and the arc labeling were implemented by Penn2Malt(http://w3.msi.vxu.se/~nivre/research/Penn2Malt.html) with the “Malt” hard-coding setting.2It was realized by Penn2Malt with the head-finding rules it provided for Chinese and the hard-coding setting.3/~strctlrn/MSTParser/MSTParser.htmlLDP1div: LDParser1 using divided models. In our experiments, the first layer instances in the training set are used to train the first layer model while the instances on all of the layers are trained for the higher layer model.The first five models are taken as the baselines in the experiments and the last three ones are the proposed parsers to be compared.The results are evaluated by the unlabeled attachment score (UAS), labeled attachment score (LAS), root accuracy (RA) and complete match (CM) according to Nivre and Scholz (2004) except that RA is the proportion of sentences in which the root word is correctly identified. All the metrics are calculated excluding the punctuations besides CM. We also present the detailed comparisons with the baselines in aspects of the computational complexity and the testing time (the CPU time). All the experiments were done on a 32-bit Intel Xeon 2.33GHz processor.3.1English ResultsIn the English experiments, all the parsers listed above except Duan07 were compared. For MaltParser, we chose the arc-eager algorithm (Nivre, 2004) and the feature set which got the best performance for English in (Hall et al., 2006) (the feature model Φ5 in their work). Hall et al. (2006) reported that the SVMs learning algorithm outperformed memory-based learning (MBL) on this feature set and could parse faster. It is the same case for Chinese. Therefore, SVMs were used for both our English and Chinese experiments. We also compared the split MaltParser which utilizes the efficient Classifiers Splitting in the experiment where the POS tag of the next input token was selected for splitting and the split threshold was 1,000. For Yamada03, the optimal feature context window size six was chosen and the dependency relation type of the child tokens was added into the feature set. The model was also trained dividedly according to the POS tag of the left target token. For MSTParser, we tried to reproduce the results in (McDonald et al., 2005) and (McDonald and Pereira, 2006) by using the 5-best projective parsing algorithm and not including punctuations in Hamming loss calculation.Considering the training cost, only the features that occur more than twice were modeled in LDParser1 and LDP1div. The combination of the children features and the combination between the word and POS features in Table 1 were also omitted.The final results are compiled in Table 2. n denotes the length of the input sentence and R is the number of the dependency types appearing in the corpus. The complexity of LDParser is a constant multiple of R2n2 according to the labeling strategies (one or two times labeling).Table 2: Parsing results on the English testing setParser UAS (%)LAS (%)RA (%)CM (%)Complexity Testing timeMaltParser MaltParser (split) Yamada03 (split) MSTParser1 MSTParser2 LDParser2 LDParser1LDP1div 89.6889.5289.5991.0391.7288.6089.1689.6888.4888.1988.7289.7890.4687.3487.9188.4384.7384.8185.1194.2194.4187.9688.7089.1633.6933.7734.1535.7239.5331.1332.6233.90O(n)O(n)O(n2)O(n3)O(n3)O(R2n2)O(R2n2)O(R2n2)2hour 46min10min 20sec20min 8sec6min 58sec9min 44sec1min 18sec2min 6sec1min 58secAmong the three proposed parsers, LDParser1 outperforms LDParser2 and LDP1div is the best one. Concerning the terms for parent-prediction accuracies and sentence complete matching, the LDParsers perform similarly to the transition-based models but exceed them more in root accuracy. Thanks to the global search over the whole dependency tree the graph-based models realized by MSTParser gain the best performance among the competitors on the English dataset. However, considering the parsing efficiency, the LDParsers are quite competitive. They havelower complexity than graph-based models and accordingly parse faster than them under the current implementations in projective case. Transition-based models can be implemented in linear time but SVMs which have been proved to achieve the highest performance in parser learning (Cheng et al., 2005; Wang et al., 2006) are not regarded as fast algorithms especially when the number of classes is large. The Classifier Splitting heuristic strategy and SVM speeding up methods (Goldberg and Elhadad, 2008) are gold choices to accelerate these implementations. However, even considering these cases, the parsing speed of the proposed LDParsers (up to 480 English words per second) is still desirable. Moreover, the speed boosting of SVMs is usually accompanied with the decrease of the accuracies or more memory consumption.3.2Chinese ResultsWe compared LDParser (LDP1div) with MaltParser, Yamada03, MSTParser and Duan07 in the Chinese experiments. Arc-standard algorithm (Nivre, 2004) is adopted in MaltParser because the experiments on the development set revealed that it got a higher performance than the arc-eager one. We also used the best Φ5 feature set in Hall et al. (2006) for Chinese and the setting of classifier splitting was kept the same as what it was for English. So were the feature model and splitting for Yamada03. All the settings for Chinese experiments of MSTParser were not changed from English ones except the 1-best parse set size. The results on the development set indicated that the k-best (k>1) models did not surpass the 1-best one remarkably.Only the features that appear more than once were utilized in LDP1div. Table 3 illustrates the parsing accuracies and speeds.Table 3: Parsing results on the Chinese testing set. The complexity of Duan07 is O(BKn2), where B is the beam size of beam-search algorithm and K is the number of action steps in PAPM (Duan et al., 2007) Parser UAS (%)LAS (%)RA (%)CM (%)Testing timeMaltParser (split) Yamada03 (split) MSTParser1 MSTParser2 Duan07LDP1div 83.8283.9183.3985.2384.3883.4482.1582.4481.7583.4782.9481.8973.5470.3870.7675.7071.2870.2932.5531.3226.3031.8132.1729.6622min 42sec27min10min 28sec15min 40sec9hour 57min1min 53secThe scores in Table 3 imply that LDParser is comparable to first-order MSTParser for Chinese parsing and a little weaker than transition-based approaches. The reason is that the transition-based models are more suitable for Chinese parsing than English because of the richer feature representations. This is also the reason why LDParser catches up with MSTParser on Chinese dataset. The optimal sub-graphs are delivered deterministically between the layers in LDParser which makes the parser be able to use the dependency graph pre-built. Duan07 which added global search to Yamada03 obtains further better performance.4Similar to the experiments for English，LDParser spends the shortest time. It parses Chinese sentences about 450 words per second. Moreover, the gaps between the speeds of LDParser and others’ consistently increase. For example, LDP1div is about 8 times faster than MSTParser2 and 15 times faster than split MaltParser while it was both 5 times faster in the English experiments. We think it is partially due to the character encoding mechanism in the Java implementation of MaltParser and MSTParser. Another reason is that the average sentence length of the Chinese testing set is 26.4 words, which is longer than that of English (23.5). Profiting from the layer-based search and sequence typed analysis, LDParser handles long 4The rank of the parsers under the metrics of parsing accuracies in Table 3 is not quite the same as what was in Duan et al. (2007). It is because the dependency structures of the data were differently converted in our experiments.sentences more efficiently. The global search of the transitions adopted in Duan07 makes the parser the most laggard one.3.3Additional ResultsTo further study the character of the layer-based parser, we present two additional results in this section. Table 4 illustrates the unlabeled attachment scores (UAS) of different dependency lengths in the English parsing experiment. The dependencies are calculated separately according to their length, equal to 1 (the neighboring relations), shorter than 3 or longer than 3. The threshold is chosen in terms of the average dependency length of the corpus which is 3.28.Table 4: Unlabeled attachment scores of different dependency lengths on the English datasetParser =1 ≤ 3 > 3MaltParser MSTParser2 LDP1div 94.2494.6794.5693.0993.5993.0773.9283.2374.53The moderate behavior of LDParser in neighboring attachment accuracy demonstrates that the globally optimal sequence labeling is competent for neighboring relation parsing compared with the tree-typed hierarchical ones. It even exceeds the transition-based parser. For the long dependencies, LDParser also does well than the transition-based one which verifies that the global search inside the parsing layer lightens the error propagation in transition-based models.By keeping partial parsing history through factoring over adjacent edge pairs of the dependency tree, the second-order MSTParser performs the best both for short and long dependencies. Making use of the pre-built structures, LDParser achieved a similar performance as MSTParser for short dependencies but gets worse for long ones. It is because LDParser is still a deterministic model in nature, the error propagation is unavoidable when the dependencies grow long. Another reason is that the higher layers are not modeled separately from each other in the current LDParser and it depresses the disambiguation ability of the model for higher layer parsing.We further examined the behaviors of the parsers on long sentences. 171 sentences with more than 40 words in the English testing set were tested and the results are listed in Table 5. The percentages represent the decrease of the speed when parsing the long sentences. Taking both the dependency accuracy and root accuracy into account, LDParser is almost the same as MaltParser. Although MSTParser is still the best, the parsing speed has dropped a lot (57%) when sentences grow long. Contrarily, there is only 8% slower for LDParser to parse these sentences which further implies that the layer-based approach is not sensitive to the length of the sentences and can be more efficient for long sentences than other parsers compared.Table 5: Results for sentences longer than 40 words. 8,019 words were analyzed in the experiment.Parser UAS RA Testing timeMaltParser (split) MSTParser2 LDP1div 87.3489.8486.5871.9294.7478.951min 45sec (17%)3min 11sec (57%)18sec (8%)4Related WorkActually, as a bottom-up framework the proposed approach is a little similar to the model proposed by Yamada and Matsumoto (2003) which employed a shift-reduce algorithm with multiple passes over the input. The transitions in this model are greedily selected at each parser state, i.e., configuration, from the left to the right during the parsing pass. To remove the greedy properties in the transition-based models, Johansson and Nugues (2007) and Duan et al. (2007) added a global search over the transition sequences. Our approach also uses multiple passes。

a r X i v :c o n d -m a t /9405029v 1 10 M a y 1994Propagation and Extinction in Branching Annihilating Random WalksD.ben-Avraham †,F.Leyvraz ‡,and S.Redner ⋆†Clarkson Institute for Statistical Physics (CISP)and Department of Physics,Clarkson University,Potsdam,NY 13699-5820‡Instituto de Fisica,Laboratorio de Cuernavaca,UNAM Apdo Postal 20-364,01000Mexico D.F.,MEXICO⋆Center for Polymer Studies and Department of PhysicsBoston University,Boston,MA 02215We investigate the temporal evolution and spatial propagation of branching annihilating random walks in one dimension.Depending on the branching and annihilation rates,a few-particle initial state can evolve to a propagating finite density wave,or extinction may occur,in which the number of particles vanishes in the long-time limit.The number parity conserving case where 2-offspring are produced in each branching event can be solved ex-actly for unit reaction probability,from which qualitative features of the transition between propagation and extinction,as well as intriguing parity-specific effects are elucidated.An approximate analysis is developed to treat this transition for general BAW processes.A scaling description suggests that the critical exponents which describe the vanishing of the particle density at the transition are unrelated to those of conventional models,such as Reggeon Field Theory.P.A.C.S.Numbers:02.50.+s,05.40.+j,82.20.-w1:Introduction In the branching annihilating random walk (BAW),a single random walk branches at somespecified rate and two random walkers annihilate at another rate when they meet.1−4As a function of these rates,the number of random walkers may grow without bound,reach a finite limiting number,or vanish asymptotically.Our goal,in this paper,is to determine some of the long-time properties of this BAW process.We are particularly interested in understanding the kinetics and density distribution when the initial state consists of a small number of localized particles.Interest in this process has several motivations.First,considerable theoretical ef-fort has been devoted to establishing the existence of and quantifying the non-equilibrium phase transition between “propagation”and “extinction”for a variety of interacting par-ticle systems 5which are closely related to BAWs.Here the term extinction refers to the situation where annihilation dominates over branching and an initially localized popula-tion of particles ultimately disappears.In the complementary case,branching dominates over annihilation and an initially localized population evolves into a propagating wavefront which advances into the otherwise empty system.Typical examples of these phenomenainclude the contact process,6Schl¨o gl models,7as well as directed percolation and Reggeon field theory.8The propagation phenomenon is also a discrete realization of the“Fisher wave”,9,10which describes the continuum dynamics of an initially localized single-species population whose evolution is influenced by diffusion,as well as by both(linear)birth and(quadratic)death mechanisms.The relation between the continuum description of the Fisher wave and the corresponding BAW is tenuous and comprehensive investigations of discrete BAW models would be helpful to understand better the relation with the con-tinuum counterpart.Second,there is a direct correspondence between the2-offspring BAW and the interface dynamics in the reaction-limited monomer-monomer surface re-action model in one dimension.11,12For the surface reaction,a lattice isfilled with A and B particles and the ensuing dynamics is defined by randomly and repeatedly selecting a nearest-neighbor AB pair and changing it to either AA,BB,AB,or BA at specified rates.The dynamics of AB interfaces is identical to that of the individual particles in the 2-offspring BAW.In addition to connections with various non-equilibrium systems,the BAW model is amenable to theoretical analysis.Somewhat surprisingly,wefind that the transition between propagation and extinction is controlled by detailed features of the underlying discrete process.In particular,the exact solution of the2-offspring BAW model in one dimension reveals that propagation occurs only for infinite branching rate and the parity of the initial number of particles essentially influences the long-time kinetics.In the next section,we outline several basic facts about the BAW process.The general conditions which lead to propagation or extinction are discussed.In Section3,we present an exact solution for the evolution of the2-offspring BAW process A→3A and2A→0, in the case where the probability of reaction when two particles meet,k,is unity.For this case,extinction is paradoxically found to occur for all non-zero values of the diffusion rate. When there is only branching,a behavior intermediate to propagation and extinction occurs.In Section4,we present an approximate description for the transition between propagation and extinction by solving a truncated hierarchy of rate equations for multi-particle correlation functions.The primary result of this treatment is that the transition emerges naturally at the next level of approximation beyond mean-field theory.In Section 5,we present numerical simulation results to support our various theoretical predictions, and then conclude in Section6.2:Models and PhenomenologyWe define the branching annihilating random walk(BAW)on a lattice as follows.A particle is picked at random.It can undergo either nearest-neighbor diffusion or it may branch with respective rates D and r.In a diffusion attempt,a random direction is picked and the particle moves to the target site if it is unoccupied.If the target site is occupied,then annihilation of the incident and target particles occurs with probability k.Otherwise, the incident particle remains at its original position.The details of the branching step depend on the number of offspring produced.In the2-offspring BAW,if a branching attempt is selected(at rate r),then branching to the two nearest-neighbor sites occurs with probability1if both neighbors are empty.If one or both neighbors are occupied,then branching to both neighbors occurs with probability k.In this branching,if a newly created particle is placed on a previously occupied site,then both particles are removed. An analogous procedure is employed for the1-offspring ly,in a branching event,if the target is occupied,then branching occurs with probability k.This branching entails immediate annihilation of the newly-created and target particles.While these microscopic rules are somewhat involved when both branching and afinite reaction rate are operative,any reasonable discrete realization of the continuum process is anticipated to give qualitatively similar results.(An exception is the case of parallel dynamics,as opposed to the serial updating assumed here.)Since D and r have dimensions of inverse time and k is dimensionless,two dimensionless parameters which characterize the system are D/r and k.As mentioned above,the BAW process with an initially localized particle population generically exhibits a transition from extinction(which is of one of two types;see below) to propagation.However,this transition appears to be at odds with a standard mean-field treatment of the model.Indeed,such a description of the BAW leads to the reaction-diffusion equation∂ρ( r,t)aD,whenever a>0. Since a is implicitly positive in this description,there appears to be no mechanism for extinction.To generate a transition to extinction,a mechanism which changes the sign of a is needed.Such a mechanism,however,is easily realizable.At low densities,mean-field theory erroneously postulates that the encounter probability varies as the square of the particle density,since all spatial correlations are neglected.However,a more complete treatment of correlations leads to an encounter probability which is linear in the particle density in the low density limit.This follow from the exact reaction-diffusion equation∂ρ( r,t)a exact/D,any reasonable theory must giveρ2( 0,t)≃ ρ + ρ 2because the birth mechanism produces nearby pairs of particles. Thus,extinction may arise in low spatial dimension because branching and annihilation terms are of comparable magnitude in Eq.(2.1);in particular,the sign of the coarse-grained branching rate a≡a exact−b exact may change as external parameters are varied.In the following,we shall use r for the branching rate and k for the annihilation probability.In numerical simulations,a transition from propagation to extinction generically oc-curs as the ratio of diffusion to branching increases(Fig.1).As the annihilation probability (equivalent to the reaction rate)increases,the extinction transition is shifted to lower val-ues of D/r.A crucial determinant for extinction is the“parity”of the branching event. For example,an even-offspring BAW process which starts with an odd number of particlescannot exhibit true extinction.Rather,the number of particles remains bounded when D/r is large.Thus extinction comes in two versions:In non-parity preserving models (odd-offspring BAW),or in parity preserving models with an even number of particles initially,there is true extinction,whereas in the complementary cases,the particle number remains bounded.The special case of the2-offspring BAW in the limits of pure branching,D/r=0, and also complete reaction,k=1,deserves emphasis because of its peculiar and easily visualized features(Fig.2).The crucial aspect of this situation is that a pair of nearest-neighbor particles diffuses rigidly under the action of the branching process.Through the branching process,a single initial particle spawns rigid pairs at afinite rate which then diffuse freely.In this situation,the number of particles grows as√N (P j+1,k(t)+P j,k+1(t)+P j,k−1(t)+P j−1,k(t)−4P j,k(t)),(3.1)while for j=k,P j,j(t+1/N)−P j,j(t)=D0+r0NP j,j(t).(3.2)Thefirst equation can be extended to include j=k by introducing the boundary condition,(D0+r0)(P j,j−1(t)+P j+1,j(t))=2r0P j,j(t).(3.3) Here we have taken the time increment for an individual birth or diffusion event to be of the order of the inverse system size so that each particle will typically be updated once in a unit time step.These equations can be simplified by transforming to the coordinates x=j+k and y=j−k.After taking continuous time derivatives,onefinds,for y≥1,˙Px,y(t)=(D0+r0)(P x+1,y+1(t)+P x+1,y−1(t)+P x−1,y+1(t)+P x−1,y−1(t)−4P x,y(t)),(3.4) while for y=0one has,˙Px,0(t)=(D0+r0)(P x+1,1(t)+P x−1,1(t)−2P x,0(t))−2D0P x,0(t).(3.5) These two equations can again be unified by imposing the boundary condition,(D0+r0)(P x+1,−1(t)+P x−1,−1(t)−2P x,0(t))=−2D0P x,0(t).(3.6) To obtain a non-singular continuum limit for these master equations,(D0+r0)must be of order(∆x)−2,whereas from the boundary condition,(D0+r0)/D0should be of order (∆x)−1.These two restrictions,in turn,imply that r0∼(∆x)−2and D0∼(∆x)−1.That is,an infinitely large microscopic birth rate is needed to counteract the instant recom-bination of newly-formed pairs in the continuum limit,thus ensuring afinite continuum birth rate.In the following,we simply replace the microscopic combination D0+r0by r0—since r0is infinitely larger than D0—and then take the continuum limit.Denoting the continuum limits of(∆x)D0and(∆x)2r0as D and r,respectively,the equation of motion for the interval occupation probability P(x,y)becomes∂P(x,y)∂y|y=0=D(2π)2e ik1(x−x0)−r(k21+k22)t e ik2(y−y0)+k2+iD/r4πDrt3/2exp{−(x−x0)2+y20which arises by neglecting the dimensionless “radiation”length r/D compared to the diffusion length√(D/r )√rt .Thus the total number of particles,N (t )≡ c (x,t )dx →2r/D as t →∞when one particle is initially present,and N (t )∼(rt )−1/2when two particles are initially present.Generally,an initially localized group with an odd number of particles leads to a finite number of particles as t →∞,while a localized initial population with an even number of particles leads to N (t )∝t −1/2.If there is no diffusion,then Eq.(3.8)reduces to a Neumann boundary condition.With this simplification,one finds the following density distribution for a single particle initial condition,after several straightforward steps and without any approximations other than those involved in the continuum limit,c (x,t )=∞0dy 0 y 0−y 0dx 0P (x,0;x 0,y 0;t ),=12√t ,intermediate to the limiting casesof Fisher wave propagation,where the particle number grows linearly in t ,and extinction,where the particle number either remains constant (through parity effects)or else decays as t −1/2.Another interesting feature of the distribution is that it qualitatively resembles a Gaussian.This can be made plausible by consideration of the underlying discrete process.When there is only branching,it is easy to verify that a nearest-neighbor particle pair propagates diffusively as a rigid unit (Fig.2).When two such soliton-like excitations meet,they merely “scatter”without any change in their form.As a single initial particle evolves,there is production of pairs at a finite rate which then diffuse freely within a region of length√also provides a basis for a scaling description of this transition.For the total number of particles,we make the following scaling ansatz,N(t)∼tβΦ(ǫtα),(3.13) whereǫ(k)is the deviation of D/r from its critical value.This parameter will be considered positive in the region where propagation occurs and negative otherwise.From the fact that N(t)grows linearly with time forǫ>0and decays as t−1/2otherwise(for the more generic case of parity non-conserving dynamics or an even number of initial particles and parity conserving dynamics),one obtainsΦ(x)∼x(1−β)/α(x>0),∼|x|−(1+2β)/(2α)(x<0).(3.14) Since the exact solution shows that N(t)remains bounded as t→∞forǫ=0,βmust equal0.(As above,we consider the case where parity is not conserved,or that the initial number of particles is even for parity conserving dynamics).Further,for small values of√D/r,this ratio appears in the exact solution in combinations of the form Dt)−1is predicted,in agreement with Eq.(3.11).4:Factorization of the Multiparticle Rate EquationsAs mentioned in Section2,the single-particle reaction-diffusion equation cannot ac-count for the transition between extinction and propagation unless there is a sign change in the coefficient of the linear term in the concentration.Although such a sign change can be justified heuristically,it is worthwhile to present a systematic continuum approach which leads to this mechanism.Our approximate treatment of the multiparticle rate equa-tions accomplishes this task.This approach also has the advantage that it can be applied straightforwardly to different microscopic branching and reaction mechanisms.In contrast, while the exact solution of the last section provides a complete description of the transition in a special case,this method is neither generalizable nor physically intuitive.Given the nature of the approximation in the multiparticle rate equations,we anticipate that our results will not depend quantitatively on the number of offspring in a branching event (except for parity-specific features),but rather,will be generic to the transition between extinction and propagation.For simplicity,we study the1-offspring BAW in one dimension which can be repre-sented asA→2A A+A→0.(4.1) It should be mentioned that the1-offspring BAW with a unit reaction probability has been investigated by rigorous mathematical techniques.1These approaches have established theexistence of a transition between propagation and extinction and provided weak bounds for the critical value of D/r.Our approximate method locates the transition for all val-ues of the reaction probability.Wefirst determine thefirst two in the hierarchy of rateequations for the multiparticle densities in this process.These equations will be closed by factorizing three-particle densities as products of two-particle densities.To write thehierarchy of rate equations,defineρ{k}as the probability that the set{k}is occupied.Thusρ0is the probability that site0is occupied,ρ0,i is the probability that sites0and i are simultaneously occupied,ρ0,i,j is the probability that0,i,and j are simultaneouslyoccupied,etc..The rate equation forρ0is found by enumerating all configurations inwhich an elemental event changes the occupancy of site0.For example,if site0is empty and site1is occupied(which occurs with probabilityρ0−ρ0,1),then by either branching or hopping to the left,site0can become occupied.There is a similar contribution if site−1is occupied.Similarly,there are three elemental events which lead to a decrease in the occupancy of site0,as illustrated in Fig.4(a).Summing these contributions,leads to therate equation˙ρ0=rρ0−[r+(2D+r)k]ρ0,1(4.2a) Similarly,the rate equation forρ0,1is obtained by enumerating all3-site configurations for which an elemental event changes the occupancy of sites0or1.These configurations and the associated rates of the processes which changeρ0,1are shown in Fig.4(b)and lead to˙ρ0,1=rρ0−(D+r)(1+k)ρ0,1+(D+r)ρ0,2−[r+(2D+r)k]ρ0,1,2(4.2b) The equations forρ0,i for i≥2are obtained similarly(Fig.4(c)),˙ρ0,i=(D+r)ρ0,i−1−2Dρ0,i+(D+r)ρ0,i+1−[r+(2D+r)k](ρ0,i,i+1+ρ0,i−1,i).(4.2c) To be soluble,these exact equations must now be closed by a suitable truncation.A standard approach is to truncate the equations at the2-particle level by the ansatzρ0,1,i+1=ρ0,1ρ1,i+1/ρ0=ρ0,1ρ0,i/ρ0.(4.3) While heuristic,this approximation yields reasonable results and turns out to be simpler than the standard Kirkwood truncation.To establish the existence of propagation,we investigate the existence of a non-zero time-independent solution to the truncated rate equations.The transition between propa-gation and extinction is then identified by the locus where this steady-state density vanishes as k and D/r are varied.To streamline the notation,let x≡D/r andρ(i)≡ρ0,i/ρ0.The steady-state rate equations reduce to1−[1+(1+2x)k]ρ(1)=0,1−(1+x)(1+k)ρ(1)+(1+x)ρ(2)−[1+(1+2x)k]ρ(1)2=0,(1+x)ρ(i−1)−2xρ(i)+(1+x)ρ(i+1)−[1+(1+2x)k]ρ(1)(ρ(i)+ρ(i−1))=0.(4.4)From thefirst of these equations,1ρ(1)=Using this in the second equation then gives1+x−xkρ(2)=1+k+2xk x.(4.8)1+k+2xkHereρ=lim i→∞ρ(i)is just the equilibrium single-particle density.Thus a positive solution forρexists only when x<1/k,corresponding to the propagating phase of the 1-offspring BAW.When x=x c=1/k,the equilibrium density vanishes;this defines the phase boundary between propagation and extinction.Note further that the decay of the equilibrium concentration is linear inǫasǫ≡x−x c→0+.If the scaling predictions of the preceding section are correct,the particle number should grow asǫ2t,forǫ→0−.On the other hand,the total particle number is the particle density multiplied by the width of the propagating wave.This latter quantity should grow asǫt,that is,the front velocity goes linearly to zero as the critical value of x is approached.5:Numerical SimulationsTo confirm the above analytical results and visualize the evolution of the system,we have performed numerical simulations of the1-and2-offspring BAW in one dimension according to the rules outlined in Section2.For concreteness,we havefixed the hopping rate to be unity so that the variables in the simulation are the branching rate r and the reaction probability k.First consider the1-offspring BAW for which it is known1,2that a transition between propagation and extinction occurs for a non-zero value of D/r.Starting with a single particle at the origin and with the parameterizations of our lattice model,the transition occurs at x≡D/r∼=1/.89.For slightly larger values of x,the number of particles initially increases,but ultimately decays to zero.For example,for x=1/.88,the average number of particles gradually increases to4.8for t≃400,but then decreases to0for longer times.By t=1.523≃11223,only about.1%of the initial configurations are still active.For400<∼t<∼10,000,the spatial distribution of the ensemble of surviving particles appears visually to be well approximated by a Gaussian(Fig.5(a)).In this time range, the reduced moments of the spatial distribution are m4≡ x4 / x2 2∼=2.8–3.02and m6≡ x6 / x2 3∼=13.0–13.9,while for a Gaussian,the corresponding values are m4=3 and m6=15.This behavior suggests that the effect of branching is irrelevant in the scalingsense,and that the spatial evolution of the1-offspring BAW coincides with that of a single purely random walk.On the other hand for x=1/.95,the simulations clearly show that a Fisher-like wavefront forms which then propagates at afinite velocity(Fig.5(b)).For the2-offspring BAW,we have approximately mapped out the phase boundary between extinction and propagation(Fig.1(b)).Simulations clearly indicate that the phase boundary intersects the line k=1at D/r=0.Thus for k=1there is extinction for all D/r except when D/r=0.At this special point,our exact solution showed that√the number of particles grows asAcknowledgementsWe gratefully acknowledge DGAPA project IN100491,and project#903922from CONACYT(FL)and ARO grant DAAL04-93-G-0021and NSF grant INT-8815438(SR), for partialfinancial support of this research.References1.M.Bramson and L.Gray,Z.Wahrsch.verw.Gebiete68,447(1985).2.A.Sudbury,Ann.Probab.18,581(1990).3.H.Takayasu and A.Yu.Tretyakov,Phys.Rev.Lett.68,3060(1992).4.I.Jensen,Phys.Rev.Lett.70,1465(1993).5.For general reviews on interacting particle systems,see e.g.,T.M.Liggett,InteractingParticle Systems,(Springer-Verlag,New York,1985);R.Durrett,Lecture Notes on Particle Systems and Percolation,(Wadsworth,Pacific Grove,CA,1988).6.T.E.Harris,Ann.Probab.2,969(1974).7.F.Schl¨o gl,Z.Phys.B253,147(1972);P.Grassberger and A.de la Torre,Ann.Phys.(N.Y.)122,373(1979).8.R.C.Brower,M.A.Furman,and M.Moshe,Phys.Lett.B76,231(1978);J.L.Cardy and R.L.Sugar,J.Phys.A13,L423(1980).9.R.A.Fisher,Ann.Eugenics7,353(1937);A.Kolmogoroff,I.Petrovsky and N.Piscounoff,Moscow Univ.Bull.Math.1,1(1937).10.J.D.Murray,Mathematical Biology,(Springer-Verlag,Berlin,1989).11.P.Krapivsky,Phys.Rev.A45,1067(1992);J.Phys.A25,5831(1992).12.H.Takayasu and H.Inui,J.Phys.A25,L585(1992).Figure CaptionsFigure1.Phase diagram of the BAW process as a function of the reaction probability k and the ratio of the diffusion to branching rate D/r.In(a)the phase diagram appropriatefor BAW processes with non-parity conserving branching is indicated.This behaviorshould be taken as representative of the continuum limit of the BAW process.In(b),the phase diagram for the2-offspring BAW is schematically indicated.In this case,the nature of the“extinct”phase depends on whether the initial number of particlesis even or odd.Figure2.Illustration of the space-time evolution of the2-offspring BAW in one dimension in the pure branching limit,D/r=0,and k=1.In this case,an isolated nearest-neighborpair of particles remains bound and executes a simple random walk.The particlewhich branches at the next step is indicated by the open circle.Figure3.Enumeration of the processes which can change the parity of the number of particles contained in the interval[j,k]under the action of the2-offspring BAW.Shown arethe various microscopic events and their corresponding statistical weights.The casesof(a)k>j and(b)k=j are somewhat different and therefore need to be treatedseparately.Figure4.Enumeration of the processes which can change the occupancy of a selected set of sites in the1-offspring BAW.In(a)the set consists of a single site,ρ0,while in(b)theset consists of2adjacent sites,ρ0,1.In(c),the set consists of2non-adjacent sites,ρ0,i.The signs to the left denote whether a process increases(+)or decreases(−)ρ.The rate of the processes(and their kind)is indicated to the right.Solid and emptycircles denote occupied and empty sites,respectively.Figure5.Spatial distribution for the1-offspring BAW in one dimension at t=1.519∼=2217.The cases shown are:(a)25000configurations at D/r=1/.88,where the survivalprobability has decreased to approximately0.11,and(b)200configurations at D/r=1/.95,where the survival probability has saturated at approximately0.81.。

㊀第40卷㊀第9期2021年9月中国材料进展MATERIALS CHINAVol.40㊀No.9Sep.2021收稿日期:2021-05-23㊀㊀修回日期:2021-08-09基金项目:国家自然科学基金面上项目(51973039,21774023)第一作者:丁㊀宁,女,1996年生,硕士研究生通讯作者:郭㊀佳,男,1979年生,教授,博士生导师,Email:guojia@DOI :10.7502/j.issn.1674-3962.202105024离子型共价有机框架材料的研究进展丁㊀宁,郭㊀佳(复旦大学高分子科学系聚合物分子工程国家重点实验室,上海200438)摘㊀要:离子型共价有机框架(ionic covalent organic frameworks,iCOFs)材料是一种框架或孔道上带有电荷的共价有机框架(COF)材料,不仅具有比表面积大㊁长程有序㊁易于功能化㊁可设计性强等传统中性COF 材料的优点,而且框架上带有的离子基团作为作用位点,结合相反电荷的分子,赋予材料选择性吸附㊁离子传导等功能,此外,其层间的静电排斥作用降低了剥离难度,因此更容易得到COF 纳米片层,这些独特的性质吸引了广大研究人员的关注㊂近年来,研究人员在iCOFs 的单体种类㊁反应类型㊁拓扑结构和制备方法等方面开展了深入研究,其在分子吸附与分离㊁催化㊁质子传导㊁能源转化㊁生物医用等领域表现出独特的优势和重要的应用价值㊂从材料的设计㊁合成方法和应用研究出发,综述了iCOFs 近年来的研究进展和发展现状,并对其研究前景进行展望㊂关键词:共价有机框架;离子化;离子型共价有机框架;多孔材料;结晶中图分类号:TQ317;TQ028;O621.25+1㊀㊀文献标识码:A㊀㊀文章编号:1674-3962(2021)09-0685-12Research Progress of Ionic Covalent Organic FrameworksDING Ning,GUO Jia(State Key Laboratory of Molecular Engineering of Polymers,Department of Macromolecular Science,Fudan University,Shanghai 200438,China)Abstract :Ionic covalent organic frameworks (iCOFs)are an emerging subclass of COFs which contain either chargedskeleton or charged pore interface.Besides the intrinsic advantages of neutral COFs such as large specific surface area,structural tunability and designability,the ionic moieties of iCOFs act as interaction sites to bind with the oppositely charged guests,thus endows iCOFs with functionality such as selective adsorption and ions conduction.The ionic repulsion between layers caused an easier exfoliation of iCOFs into ionic covalent organic nanosheets (iCON).These promising features have attracted considerable interests.In recent years,the building blocks,reaction types,topology structures and synthetic strate-gies of iCOFs have been studied extensively,paving the way of various applications such as molecular adsorption /separation,catalysis,energy conversion,proton conduction and biomedical engineering.In this review,we summarize the research pro-gress of iCOFs based on design,synthesis and applications.The current challenge and future directions for the research of iCOFs are discussed as well.Key words :covalent organic frameworks;ionization;ionic covalent organic frameworks;porous materials;crystallinity1㊀前㊀言共价有机框架(covalent organic frameworks,COFs)材料是一类由共价键连接的结晶型有机多孔聚合物,具有稳定的骨架和开放而规则的孔道结构㊂自从2005年Yaghi 课题组[1]利用动态共价化学的原理和对称的有机构筑单元合成第一例COF 以来,COF 的构筑单元㊁连接形式㊁合成方法㊁拓扑结构得到了不断扩展,关于COF 构效关系的研究不断深入,因其孔径可调㊁合成方法多样㊁易于功能化㊁可设计性强等优点,近年来被广泛应用于分子吸附与分离[2-16]㊁能源转化[17-30]㊁质子传导[31-35]㊁催化[36-50]㊁生物医用[51,52]㊁荧光检测[53-55]等领域,展现出十分广阔的应用前景㊂从化学组成角度,COF 材料由C,H,O,N,B 等轻元素组成,和无机材料以及金属材料相比具有极低的密度㊂与金属有机框架材料相比,COF 材料由共价键连接,具有更好的热稳定性和化学稳定性㊂此外,COF 材料博看网 . All Rights Reserved.中国材料进展第40卷不仅化学框架可以设计,孔径㊁形貌㊁孔道表面环境都有很高的设计自由度,因此,经过十几年的发展,COF 材料的研究仍然在不断深入的探索中㊂目前的报道主要集中在非离子型的COF 框架,而离子型COF 相对较少,也相应限制了COF 材料在某些特定领域的价值体现㊂离子型共价有机框架(ionic covalent organic frame-work,iCOF)不仅保持了COF 结构特有的原子周期性和骨架多孔性,而且在框架或侧基上有带电荷的基团,从而能够与特定结构相互协同展现出功能增强的特点㊂在材料设计和合成中,iCOF 也保持了非离子型COF 的设计理念,具有一定的电荷密度和分子间强的相互作用,不仅保留了COF 可设计性强㊁结构稳定㊁孔道规则的特点,还将静电相互作用引入框架,进一步扩充了COF 的类型和应用领域,通过改变离子的种类也可以对iCOF 的比表面积㊁孔径㊁孔体积㊁性质进行调控㊂本文从材料的设计㊁制备及应用等角度综述了近年来离子型共价有机框架材料的研究进展,并对其目前存在的挑战和应用前景进行展望㊂2㊀合成方法离子型COF 材料的制备方法可分为自下而上法和后修饰法两种㊂通过自下而上的方法,一方面可以设计离子型的构筑基元,如溴化乙啶(EB)㊁2,5-二氨基苯磺酸等(图1),采用典型的动态化学反应,直接合成iCOF,这一过程往往需要结构的自适应调整来平衡静电作用,从而获得具有典型重叠堆积的二维COF 结构,或是多重穿插结构的三维COF,结构中的离子可以在框架上或是悬挂在侧基上;另一方面,采用非离子构筑基元,通过形成螺硼酸酯键㊁方酸菁键等具备离子的连接键,也可以构筑离子型的COF 框架㊂然而,在大多数情况下,带电荷的构筑基元往往不利于形成高结晶的COF 材料[3],这可能是框架上吸引或排斥的电荷作用影响了特定的结构排列,也可能是离子化构筑基元的相互作用改变了反应动力学㊂此外,采用后修饰法在COF 框架上引入离子基团,即通过构筑基元上的反应位点进行功能化,这种方法具有较好的可控性,能兼具结构的结晶性和离子化特性,可根据使用要求来灵活调控㊂相比于自下而上的设计合成,后修饰法具有一定的普适性,然而如何同时获得高结晶性和高离子化的iCOF 结构仍具有一定的挑战㊂在二维iCOF 中,由于层与层之间存在一定的电荷排斥作用,和非离子型的COF 相比,iCOF 可以更容易地剥离成少层的离子化共价有机纳米片(ionic covalent organicnanosheet,iCON),由此,在合成iCOF 膜时,可将带有相反电荷的iCON 层层组装[14]获得堆叠紧密的膜,从而降低膜的厚度,提高膜的稳定性;将阳离子共价有机纳米片(CON)与带负电的Kevlar 层层堆叠[26],可以获得超薄的酰亚胺-COF 杂化膜,从而推动了iCOF 膜合成方法的研究和功能性的拓展㊂3㊀组成结构根据所带电荷的不同,iCOF 分为阳离子型㊁阴离子型和两性离子型㊂3.1㊀阳离子型COF阳离子型COF 通常使用带正电荷的单体与非离子型连接基元通过自下而上的方法直接合成(图2),其中常用的离子单体包括溴化乙啶㊁三氨基胍盐酸盐㊁5,5ᶄ-二氨基-2,2ᶄ-联吡啶㊁5,6-双(4-甲酰苯基)-1,3-二甲基苯并咪唑等(图1)㊂图1㊀常见离子型构筑基元Fig.1㊀Reported ionic building blocks㊀㊀溴化乙啶由于具有较好的平面性以及对离子可调的特点,常被用于各种iCOF 的设计合成㊂2016年,Zhu等[31]用带正电荷的溴化乙啶作为单体,与三醛基间苯三酚通过席夫碱反应合成了一系列阳离子型COF(EB-COF ʒX,686博看网 . All Rights Reserved.㊀第9期丁㊀宁等:离子型共价有机框架材料的研究进展X =F -,Cl -,Br -,I -)(图2a )㊂2018年,Ajayaghosh 等[53]将EB-COF 的水溶液室温下静置48h,获得自剥离的阳离子型COF 片层㊂和二维iCOF 层层堆叠的结构相比,三维iCOF 的贯穿结构可以在一定程度上减弱框架内相邻电荷之间的排斥作用,从而有助于有序结构的稳定㊂2017年,Qiu 等[4]合成了2种三维离子型iCOF(图2b),其平面单体分别为溴甲菲啶和溴化乙啶,由于框架自带离子,贯穿度从非离子型COF-320的九重贯穿变为三重贯穿㊂2016年,Banerjee 等[51]用三醛基间苯三酚与三氨基胍盐酸盐作为单体合成了自剥离的胍基iCON(图2f),对金黄色葡萄球菌和大肠杆菌均有很好的抗菌效果㊂此后,具有类似结构的胍基COF 被应用于锂离子传导[19]㊁气体分离[5]㊁污染物吸附[6,9]等领域并展现出良好的应用潜力㊂图2㊀阳离子型COF 结构:(a)溴化乙啶型iCOF [31],(b)三维溴化乙啶型iCOF [4],(c)苯并咪唑型iCOF [3],(d)联吡啶季铵盐iCOF [52],(e,f)胍基连接的iCOF [51]Fig.2㊀Structure of cationic COF:(a)ethidium bromide (EB)-based iCOF [31],(b)EB-based 3D-iCOF [4],(c)benzimidazolium-basediCOF [3],(d)quaternized 2,2ᶄ-bipyridine-based iCOF [52],(e,f)guanidinium-based iCOF [51]3.2㊀阴离子型COF阴离子型COF 主要包括螺硼酸酯为阴离子框架以及磺酸为框架侧基的2种类型(图3)㊂2015年,Zhang 等[17]用三甲基硼酸酯和二醇官能化的大环分子为单体,二甲胺或氢氧化锂为碱催化剂,通过酯交换反应合成了对离子为二甲胺阳离子(Me 2NH 2+)或Li +的螺硼酸酯阴离子iCOF-1和iCOF-2㊂2017年,Wang 等[18]通过微波辅助法使用γ-环糊精构建了螺硼酸酯连接的三维COF,具有多样的平衡阳离子,如Li +㊁HDMA +㊁H 2PPZ 2+等㊂采用自下而上法和后修饰法均可实现在COF 框架侧基悬挂磺酸基,因此在COF 组成设计上更加多样化,近年来受到较多的关注㊂2019年,Lee 等[21]使用2,5-二氨基苯磺酸为单体,合成了侧基含磺酸基的TpPa-SO 3COF,对其进行锂化可以获得含自由锂离子和锚定阴离子的TpPa-SO 3Li(图3c),表现出优异的单锂离子传导性质㊂2021年,Jiang 等[56]通过界面法直接合成了自支撑的TpPa-SO 3H COF 膜,该材料具有贯通的孔道和较高的电荷密度,在盐差能量转换方面表现出应用潜力㊂Gu 等[16]使用2,2ᶄ-联苯胺二磺酸为单体合成了磺酸功能化的TFP-BDSA COF 球(图3b),对多种阳离子染料表现出优异的吸附性㊂3.3㊀两性离子型COF目前已报道的两性离子型COF 材料主要是框架上含两性离子连接键的方酸COF 和两性离子液体后修饰的COF(图4)㊂方酸与胺类物质反应可以得到具有共振稳定两性离子特征的方酸菁结构㊂2013年,Jiang 等[36]用自下而上的方法合成了方酸菁连接的CuP-SQ COF(图4a),虽然框架中分布有大量的正负电荷,但该材料786博看网 . All Rights Reserved.中国材料进展第40卷图3㊀阴离子型COF结构:(a)螺硼酸酯型iCOF[17],(b)磺酸基型iCOF[16],(c)锂磺化的iCOF[21] Fig.3㊀Structure of anionic COF:(a)spiroborate-linked iCOF[17],(b)sulfonated iCOF[16],(c)lithium sulfonated iCOF[21]图4㊀两性离子型COF结构:(a)方酸菁连接的iCOF[36],(b)两性离子液体后修饰的iCOF[39] Fig.4㊀Structure of zwitterionic COF:(a)squaraine-linked iCOF[36],(b)ionic liquid decorated iCOF[39]仍然具有很好的结晶性和较高的比表面积,在可见光下表现出诱导产生单线态氧的能力㊂4㊀应㊀用4.1㊀分子吸附与分离非离子型COF可以通过极性相互作用和纳米孔道的物理尺寸截留效应吸附客体分子,相比而言,iCOF可通过静电相互作用结合带有相反电荷的客体分子,客体分子还可以通过离子交换的方式进入COF框架,从而实现对客体分子的高效负载,这使得iCOF可通过尺寸和电荷双重控制分子的吸附与分离,特别是对离子污染物的清除更加完全且选择性更强㊂2016年,Li等[2]以1,1ᶄ-二苯基-4,4ᶄ-二氯化联吡啶鎓(BFBP2+㊃2Cl-)为构筑基元合成了一种具有较大孔径(5.8nm)的阳离子PC-COF,对水中多种阴离子有机染料表现出较高的吸附能力(>97%),检测限度为3.2ˑ10-5mol㊃L-1㊂然而,PC-COF的结晶性并不理想,作者将其归因于BFBP2+之间较强的排斥力阻碍了层与层之间的规则堆叠㊂为改善这一问题,2017年,Jiang等[3]合成了1,3-二甲基苯并咪唑溴结构的阳离子COF,相邻层的咪唑阳离子在框架两侧交替堆叠,减少了层间电荷排斥作用,提高了材料的结晶性,比表面积为1532m2㊃g-1,对甲基橙表现出553mg㊃g-1的捕获能力;该研究也证明了此材料无法吸附中性分子和阳离子污染物,说明特定886博看网 . All Rights Reserved.㊀第9期丁㊀宁等:离子型共价有机框架材料的研究进展设计的离子型COF在吸附离子污染物时具有较强的电荷选择性㊂近两年来,越来越多的iCOF被用于吸附水溶液中的重金属污染物,如胍基连接的BT-GD Cl吸附CrO42-(200mg㊃g-1)[7]和2,4-二氯苯酚(893mg㊃g-1)[9]㊁EB-COF/Fe3O4复合微球作为磁性固相萃取吸附剂吸附C9~C12的全氟羧酸(检测限度为0.1~0.8ng㊃L-1,回收率73.9%~108.3%)[15]等㊂与之相对,共轭微孔聚合物(conjugated microporous polymer,CMP)对CrO42-㊁甲基橙的吸附容量分别为60mg㊃g-1[57]和240.1mg㊃g-1[58],均低于前文所述iCOF的吸附水平,说明COF有序的结构有利于吸附客体分子㊂此外,后修饰的策略也常常被用于制备具有吸附功能的iCOF,如咪唑酯修饰的磺酸基COF[10]对亚甲基蓝表现出2865.3mg㊃g-1的吸附效果,而类似结构的CMP最大吸附能力仅为1650mg㊃g-1[59];阳离子表面活性剂修饰的DhaTab-COF[11]用于吸附硝酸盐,最大吸附能力为108.8mg㊃g-1,相比修饰之前的非离子型COF提高了15倍㊂三维COF常常比二维COF具有更高的比表面积和更小的孔径,对其进行离子化有望获得更好的离子染料吸附效果㊂2017年,Qiu等[4]报道了首例三维iCOF并研究了其去除核废料中放射性锝元素的应用潜力,作者选用高锰酸盐作为替代高锝酸盐的模型物质,室温下20mg的该材料可以在20min内除去水中几乎100%的MnO4-(10mg)㊂此后,研究人员对iCOF作为吸附剂捕获核废料中氧阴离子的能力进行了进一步的研究㊂2019年,Wang等[8]合成了一种紫罗碱型阳离子型COF,此种COF可以选择性吸附高锝酸根(99TcO4-)以及其非放射性替代物高铼酸根(ReO4-),ReO4-具有较高的吸附量(702.4mg㊃g-1)和快速的吸附表现(1min内可以达到吸附平衡)㊂2019年,Yan等[6]以三氨基胍盐酸盐(TG Cl)为构筑基元,合成了对水稳定的阳离子型COF纳米片(DhaTG Cl),室温下对ReO4-的最大吸附容量为437mg㊃g-1,在使用更为复杂的核废水模拟样品进行测试时,约73%的ReO4-可以被有效除去,作者认为这种良好的吸附作用不仅来源于离子交换,胍基和Dha提供的多个氢键作用也对这一过程有所贡献(图5c)㊂这些研究表明iCOF在核燃料后处理方面具有较为广阔的前景㊂在COF中引入阴离子基团可以赋予材料吸附阳离子的性质㊂近日,Gu等[16]合成了一种磺酸功能化的球形COF,该材料对亚甲基蓝㊁结晶紫㊁罗丹明B等阳离子染料均表现出较好的吸附行为,最大吸附容量分别为1116,1429和1638mg㊃g-1,而对甲基橙㊁荧光素钠等阴离子染料则无吸附效果㊂除了COF粉末,研究人员也对iCOF膜材料展开了研究㊂2018年,Ma等[60]用界面法合成了一种阳离子型EB-COFʒBr纳米片,经过简单抽滤后,这些剥离的纳米片以层层自组装的方式形成连续而致密的膜,由于孔道内部带有丰富的正电荷位点,EB-COFʒBr膜可以有效截留阴离子染料,截留率高达98%,而对阳离子和非离子型的染料分子只有较弱的截留效应(图5a)㊂传统非离子型CON的层间相互作用较弱,堆叠较为松散,限制了COF膜在气体分离方面的应用㊂基于iCOF 独特的带电荷性质,研究人员发展了新的方法来控制iCOF的堆叠模式和孔径㊂2020年,Zhao等[14]使用两种带有相反电荷的iCON(TpEBr和TpPa-SO3Na),通过层层自组装法合成了超薄COF膜TpEBr@TpPa-SO3Na (图5b),由于两种COF纳米片带有相反电荷且孔径不匹配,所得的COF膜层与层之间通过强静电相互作用实现了紧密的交替堆叠,膜厚仅为41nm且具有较小的孔径,适合用于分子筛分和气体分离,H2渗透率为8.59ˑ10-7mol/(m2㊃s㊃Pa),50ħ下H2/CO2的分离因数为22.6,比单一的TpEBr或TpPa-SO3Na iCON膜高得多㊂这种交替堆叠带有相反电荷的iCON的策略为开发高效的分子筛分膜提供了新思路㊂4.2㊀离子传导锂离子二次电池被广泛应用在储能器件和各种移动设备中,其中锂离子传导材料是构筑高能量密度㊁高转化效率的锂电池的关键㊂目前锂离子电池常用的是液态电解质,具有易燃性,且容易产生锂枝晶,造成电池短路甚至起火,对电池安全造成很大威胁㊂因此,新一代的锂离子电池正逐渐过渡到全固态电池,即用固态或准固态的电解质替代传统的电解液,由于不存在电解液与锂金属接触的界面,因此从根源上避免了锂枝晶的产生,不仅可以提高电池的性能,其安全性也有了保障㊂与非晶态的聚合物材料相比,COF框架上周期性的原子分布和规则的孔道结构,为锂离子传输提供了有序通道,缩小了离子迁移的路径,有助于离子的高效传输;而iCOF上均匀分布的正电荷或负电荷更是为锂离子提供了传输位点,降低了Li+传输的能垒㊂因此,iCOF在锂离子传导方面展现出较大的优势㊂2016年,Zhang 等[17]合成了螺硼酸酯连接的阴离子型COF,Li+为平衡离子,与聚偏氟乙烯混合后,室温下Li+电导率为3.05ˑ10-5S㊃cm-1,这也是iCOF首次被用于锂离子传导㊂2017年,Feng等[18]用微波法合成了一种三维环糊精阴离子型COF,测试前使用LiPF6-EC-DMC电解液作为载锂试剂,室温下锂离子电导率为2.7ˑ10-3S㊃cm-1㊂2019年,Zhang等[20]报道了一种咪唑连接的阳离子型COF (R-Li-ImCOF,R=H,CH3,CF3),在取代基为三氟甲基986博看网 . All Rights Reserved.中国材料进展第40卷时,室温下锂离子电导率最高,可达7.2ˑ10-3S㊃cm-1㊂在这些应用研究中,通常会在测试时加入少量有机溶剂(碳酸丙烯酯㊁碳酸乙烯酯㊁碳酸二甲酯等)以提高离子传导的性能,并非真正的固态锂离子传导,此后,研究人员致力于减少溶剂的使用㊂柔性聚合物链段可以起到模拟溶剂的作用,将PEG㊁PEO等分子引入COF框架,可以在无溶剂的情况下达到较高的离子电导率㊂2019年, Feng等[61]合成了一种短链PEG负载的阳离子型COF,在不使用溶剂的情况下室温锂离子电导率为1.93ˑ10-5S㊃cm-1,120ħ下锂离子电导率为1.78ˑ10-3S㊃cm-1㊂图5㊀iCOF在分子吸附与分离方面的应用:(a)EB-COFʒBr的结构和EB-COFʒBr膜的合成路线[60],(b)TpEBr@TpPa-SO3Na膜的堆叠模式示意图[14],(c)DhaTG Cl的合成路线和ReO4-吸附动力学[6]Fig.5㊀iCOF applied in molecular adsorption and separation:(a)structure of EB-COFʒBr and synthesis of EB-COFʒBr membrane[60],(b) scheme of different stacking modes[14],(c)synthesis of DhaTG Cl and adsorption kinetics of DhaTG Cl for ReO4-[6]㊀㊀固态电解质材料的理想状态是在室温下不仅具有高锂离子电导率,而且锂离子迁移数t Li+接近1,即单锂离子传导,这也是评价锂离子电池固态电解质的重要指标㊂由于阳离子COF可以捕获锂盐中的阴离子,减弱阴离子的运动,从而使锂离子的传导比重增加,在早期的研究中,研究人员多采用对阳离子型COF进行锂化的方法来制备锂离子导体㊂2018年,Chen等[19]通过阴离子交换将双三氟甲磺酰亚胺阴离子(TFSI-)锚定在阳离子型COF框架上(图6a),在不加入溶剂的情况下Li-CON-TFSI的室温锂离子电导率为5.74ˑ10-5S㊃cm-1,活化能为0.34eV(图6c),t Li+为0.61㊂然而,阳离子型COF和锂盐阴离子的结合是离子键合(图6b),虽然可以在一定程度上降低锂盐阴离子的运动能力,提高t Li+,但效果并不理想㊂基于此,可以制备阴离子型COF作为锂离子导体,直接将锂盐阴离子通过共价键作用锚定到COF框架中,有望实现单锂离子传导㊂2019年,Lee等[21]合成了锂磺化的TpPa-SO3Li(图6d),其中磺酸根阴离子通过共价键锚定在COF的框架上(图6e),在不添加任何溶剂的情况下室温锂离子电导率为2.7ˑ10-5S㊃cm-1,活化能为0.18eV(图6f),t Li+为0.9,这也是目前报道的锂离子迁移数的最高值之一㊂096博看网 . All Rights Reserved.㊀第9期丁㊀宁等:离子型共价有机框架材料的研究进展图6㊀iCOF 在锂离子传导方面的应用:(a,b)Li-CON-TFSI 的结构和结合锂离子示意图,(c)Li-CON-TFSI 不同温度下的离子传导率[19];(d,e)TpPa-SO 3Li 的结构和传导锂离子机制示意图,(f)TpPa-SO 3Li 不同温度下的离子传导率[21]Fig.6㊀iCOF applied in lithium ions conduction:(a,b)structure of Li-CON-TFSI and schematic illustrations of ion association,(c)ionic conductivi-ty of Li-CON-TFSI at different temperature [19];(d,e)structure of TpPa-SO 3Li and conceptual illustrations of ion transport,(f)ionic con-ductivity of TpPa-SO 3Li at different temperature [21]㊀㊀基于iCOF 亲锂性的特点,近年来,研究人员进一步将iCOF 应用到二次电池的设计中㊂2019年,Chen 等[22]通过溶液法在无机固态电解质Li 6.75La 3Zr 1.75Ta 0.25O 12(LLZTO)表面生长了一层磺酸功能化的COF(sCOF)薄膜(图7a),使用熔融的锂对sCOF 薄膜进行锂化之后,无机电解质的亲锂性提高,与锂电极之间的界面电阻显著降低,组装成的全固态锂金属电池在2C 倍率下放电比容量为97mAh㊃g -1,这是因为sCOF 材料在固态电解质和锂电极之间充当界面层,形成了能够快速传输锂离子的通道(图7b)㊂2020年,Chen 等[26]合成了胍基连接的阳离子型COF 纳米片,与聚对苯二甲酰对苯二胺(Kevlar)形成的复合膜厚度仅为7.1μm 且具有良好的力学强度,室温下离子电导率为1.62ˑ10-4S㊃cm -1,组装成的全固态LiFePO 4/Li 电池循环300h 放电比容量仅衰减0.052%,可循环500h 而不会造成短路㊂除了作为固态电解质,iCOF 还被应用于锂金属电池的负极保护㊁锂硫电池正极宿主材料等方面㊂2021年,Lu 等[28]合成了一种以SiO 62-为节点㊁以蒽为连接基元的图7㊀iCOF 在锂电池和锂硫电池方面的应用:(a,b)sCOF 膜的结构及传导锂离子示意图[22],(c)ACOF 结构及电解质界面保护膜示意图[28],(d)EB-COF-Br 负载多硫化物示意图[30]Fig.7㊀iCOF applied in lithium batteries and lithium-sulfur batteries:(a,b)structure of sCOF and schematic of lithium ion conduction [22],(c)structure of COF and schematic of Li anode protection [28],(d)the sulfur loading process of EB-COF-Br [30]196博看网 . All Rights Reserved.中国材料进展第40卷阴离子型COF(ACOF),平衡离子为Li +,旋涂于锂金属负极表面(图7c),该方法不仅有效抑制了锂枝晶的生长,而且ACOF 本身的亲锂性和多孔结构有利于锂离子的快速传输㊂相比于无保护的锂金属电池,ACOF 保护的锂金属全电池(LiCoO 2|ACOF-Li)的充放电稳定性和倍率性能都得到了显著的提升,在4.5V 高电压下可循环500次,10C 倍率下放电比容量为89mAh㊃g -1㊂锂硫电池中的多硫化锂穿梭效应会导致活性物质损失㊁电池容量衰减,2020年,Zang 等[30]通过Br -和S 82-的阴离子交换反应在EB-COF 孔道内负载多硫化物(图7d),可以有效抑制Li 2S 8溶解在电解液中,与非离子型COF 组装的电池相比充放电循环性能有所提升,4C 倍率下循环300次仍可保持较高的比容量(468mAh㊃g -1)㊂此外,其他种类的iCOF 也被应用于锂硫电池中,如胍基结构的阳离子COF [27]㊁通过点击反应后修饰季铵盐的COF [29]等㊂与COF 不同,非晶态的有机多孔聚合物材料不具备有序的传导通道,故很少用于离子传导㊂4.3㊀质子传导近年来,质子交换膜在燃料电池等能源转换器件中的应用得到了广泛关注,如何提高材料的质子传导性能是研究重点之一㊂2016年,Zhu 等[31]合成了溴化乙啶型阳离子型COF(EB-COF:Br),通过离子交换将多酸磷钨酸(PW 12O 403-)掺杂到COF 框架中可以显著提高其质子传导性能(图8a),室温下质子传导率为3.32ˑ10-3S㊃cm -1,相比掺杂前的材料提高了103倍(图8b),这是由于PW 12O 403-阴离子与水分子相互作用后在孔道内形成氢键网络,为质子传导提供了高速迁移的通路㊂2020年,Chen 等[34]报道了一种将离子化COF 和有机凝胶复合来促进质子传导的策略,酮烯胺连接的COF 在LiH 处理下去质子化形成框架含 N - Li +的阳离子型COF,40ħ㊁98%相对湿度下质子传导率为2.7ˑ10-2S㊃cm -1,相比修饰前的COF 提高了107倍;将其与小分子凝胶单体吡咯/磷酸原位凝胶化得到柔性COF 薄膜,质子传导率进一步提高至0.13S㊃cm -1,组装成的离子交换膜燃料电池在50ħ下的输出功率密度可达54mW㊃cm -2㊂无定形的有机多孔聚合物材料由于没有规则的孔道结构,质子传导方面的应用报道较少㊂Nafion 是一种全氟磺酸型聚合物,在高湿度下具有优异的质子传导特性,起关键作用的是聚合物网络中的磺酸基团㊂受此启发,研究人员认为将磺酸基团引入COF 框架可以赋予其传导质子的能力㊂2017年,Ghosh等[32]对多孔有机框架PCF-1进行磺酸功能化,30ħ㊁95%相对湿度下质子传导率约0.026S㊃cm -1,相比功能化前的材料提高了130倍㊂虽然后修饰法较为方便,但这种方法得到的iCOF 无法实现离子在框架内的均匀分布㊂2020年,Jiang 等[35]使用2,5-二氨基苯磺酸为单体,通过自下而上的方法在水溶液中直接合成了磺酸功能化的iCON (NUS-9),进一步自组装形成自支撑的膜(图8c),在80ħ㊁98%相对湿度下,质子传导率高达0.38S㊃cm -1,是目前文献报道的最高值之一,即使相对图8㊀iCOF 在质子传导方面的应用:(a)EB-COF ʒBr 掺杂多酸磷钨酸示意图[31],(b)EB-COF ʒBr 和EB-COF ʒPW 12质子传导率对比[31],(c)IPC-COF 膜合成路线和传导质子示意图[35]Fig.8㊀iCOF applied in proton conduction:(a)schematic of PW 12O 403-doping in EB-COF ʒBr [31],(b)comparison of proton conductivity ofEB-COF ʒBr and EB-COF ʒPW 12[31],(c)structure and synthesis of IPC-COF membrane and schematic of proton conduction [35]296博看网 . All Rights Reserved.。