Lecture Notes in Physics EditorialBoard R.Beig,Wien,Austria J.Ehlers,Potsdam,Germany U.Frisch,Nice,France K.Hepp,Zu¨rich,Switzerland W.Hillebrandt,Garching,Germany D.Imboden,Zu¨rich,Switzerland R.L.Jaffe,Cambridge,MA,USA R.Kippenhahn,Go¨ttingen,Germany R.Lipowsky,Golm,Germany H.v.Lo¨hneysen,Karlsruhe,Germany I.Ojima,Kyoto,Japan H.A.Weidenmu¨ller,Heidelberg,Germany J.Wess,Mu¨nchen,Germany J.Zittartz,Ko¨ln,Germany 3 Berlin Heidelberg NewYork Barcelona HongKong London Milan Paris Singapore Tokyo TheEditorialPolicyforProceedings TheseriesLectureNotesinPhysicsreportsnewdevelopmentsinphysicalresearchandteaching–quickly, informally,andatahighlevel.Theproceedingstobeconsideredforpublicationinthisseriesshouldbelimited toonlyafewareasofresearch,andtheseshouldbecloselyrelatedtoeachother.Thecontributionsshouldbe ofahighstandardandshouldavoidlengthyredraftingsofpapersalreadypublishedorabouttobepublished elsewhere.Asawhole,theproceedingsshouldaimforabalancedpresentationofthethemeoftheconference includingadescriptionofthetechniquesusedandenoughmotivationforabroadreadership.Itshouldnot beassumedthatthepublishedproceedingsmustreflecttheconferenceinitsentirety.(Alistingorabstracts ofpaperspresentedatthemeetingbutnotincludedintheproceedingscouldbeaddedasanappendix.) WhenapplyingforpublicationintheseriesLectureNotesinPhysicsthevolume’seditor(s)shouldsubmit sufficientmaterialtoenabletheserieseditorsandtheirrefereestomakeafairlyaccurateevaluation(e.g.a completelistofspeakersandtitlesofpaperstobepresentedandabstracts).If,basedonthisinformation,the proceedingsare(tentatively)accepted,thevolume’seditor(s),whosename(s)willappearonthetitlepages, shouldselectthepaperssuitableforpublicationandhavethemrefereed(asforajournal)whenappropriate. Asarulediscussionswillnotbeaccepted.TheserieseditorsandSpringer-Verlagwillnormallynotinterfere withthedetailededitingexceptinfairlyobviouscasesorontechnicalmatters. Finalacceptanceisexpressedbytheserieseditorincharge,inconsultationwithSpringer-Verlagonlyafter receivingthecompletemanuscript.Itmighthelptosendacopyoftheauthors’manuscriptsinadvanceto theeditorinchargetodiscusspossiblerevisionswithhim.Asageneralrule,theserieseditorwillconfirm histentativeacceptanceifthefinalmanuscriptcorrespondstotheoriginalconceptdiscussed,ifthequalityof thecontributionmeetstherequirementsoftheseries,andifthefinalsizeofthemanuscriptdoesnotgreatly exceedthenumberofpagesoriginallyagreedupon.ThemanuscriptshouldbeforwardedtoSpringer-Verlag shortlyafterthemeeting.Incasesofextremedelay(morethansixmonthsaftertheconference)theseries editorswillcheckoncemorethetimelinessofthepapers.Therefore,thevolume’seditor(s)shouldestablish strictdeadlines,orcollectthearticlesduringtheconferenceandhavethemrevisedonthespot.Ifadelayis unavoidable,oneshouldencouragetheauthorstoupdatetheircontributionsifappropriate.Theeditorsof proceedingsarestronglyadvisedtoinformcontributorsaboutthesepointsatanearlystage. Thefinalmanuscriptshouldcontainatableofcontentsandaninformativeintroductionaccessiblealsoto readersnotparticularlyfamiliarwiththetopicoftheconference.ThecontributionsshouldbeinEnglish.The volume’seditor(s)shouldcheckthecontributionsforthecorrectuseoflanguage.AtSpringer-Verlagonlythe prefaceswillbecheckedbyacopy-editorforlanguageandstyle.Gravelinguisticortechnicalshortcomings mayleadtotherejectionofcontributionsbytheserieseditors.Aconferencereportshouldnotexceedatotal of500pages.Keepingthesizewithinthisboundshouldbeachievedbyastricterselectionofarticlesandnot byimposinganupperlimittothelengthoftheindividualpapers.Editorsreceivejointly30complimentary copiesoftheirbook.Theyareentitledtopurchasefurthercopiesoftheirbookatareducedrate.Asaruleno reprintsofindividualcontributionscanbesupplied.NoroyaltyispaidonLectureNotesinPhysicsvolumes. Commitmenttopublishismadebyletterofinterestratherthanbysigningaformalcontract.Springer-Verlag securesthecopyrightforeachvolume. TheProductionProcess Thebooksarehardbound,andthepublisherwillselectqualitypaperappropriatetotheneedsoftheauthor(s). Publicationtimeisabouttenweeks.Morethantwentyyearsofexperienceguaranteeauthorsthebestpossible service.Toreachthegoalofrapidpublicationatalowpricethetechniqueofphotographicreproductionfrom acamera-readymanuscriptwaschosen.Thisprocessshiftsthemainresponsibilityforthetechnicalquality considerablyfromthepublishertotheauthors.Wethereforeurgeallauthorsandeditorsofproceedingsto observeverycarefullytheessentialsforthepreparationofcamera-readymanuscripts,whichwewillsupplyon request.Thisappliesespeciallytothequalityoffiguresandhalftonessubmittedforpublication.Inaddition, itmightbeusefultolookatsomeofthevolumesalreadypublished.Asaspecialservice,weofferfreeof chargeLATEXandTEXmacropackagestoformatthetextaccordingtoSpringer-Verlag’squalityrequirements. Westronglyrecommendthatyoumakeuseofthisoffer,sincetheresultwillbeabookofconsiderably improvedtechnicalquality.Toavoidmistakesandtime-consumingcorrespondenceduringtheproduction periodtheconferenceeditorsshouldrequestspecialinstructionsfromthepublisherwellbeforethebeginning oftheconference.Manuscriptsnotmeetingthetechnicalstandardoftheserieswillhavetobereturnedfor improvement. ForfurtherinformationpleasecontactSpringer-Verlag,PhysicsEditorialDepartmentII,Tiergartenstrasse17, D-69121Heidelberg,Germany Serieshomepage–http://www.springer.de/phys/books/lnpp Klaus R. Mecke Dietrich Stoyan (Eds.) Statistical Physics and Spatial Statistics The Art of Analyzing and Modeling Spatial Structures and Pattern Formation 1 3 Editors KlausR.Mecke FachbereichPhysik BergischeUniversita¨tWuppertal 42097Wuppertal,Germany DietrichStoyan Institutfu¨rStochastik TUBergakademieFreiberg 09596Freiberg,Germany LibraryofCongressCataloging-in-PublicationDataappliedfor. DieDeutscheBibliothek-CIP-Einheitsaufnahme Statisticalphysicsandspatialstatistics:theartofanalyzingand modelingspatialstructuresandpatternformation/KlausR.Mecke; DietrichStoyan(ed.).-Berlin;Heidelberg;NewYork;Barcelona; HongKong;London;Milan;Paris;Singapore;Tokyo:Springer, 2000 (Lecturenotesinphysics;Vol.554) (Physicsandastronomyonlinelibrary) ISBN3-540-67750-X ISSN0075-8450 ISBN3-540-67750-XSpringer-VerlagBerlinHeidelbergNewYork Thisworkissubjecttocopyright.Allrightsarereserved,whetherthewholeorpartofthe materialisconcerned,specificallytherightsoftranslation,reprinting,reuseofillustra- tions, recitation, broadcasting, reproduction on microfilm or in any other way, and storageindatabanks.Duplicationofthispublicationorpartsthereofispermittedonly undertheprovisionsoftheGermanCopyrightLawofSeptember9,1965,initscurrent version,andpermissionforusemustalwaysbeobtainedfromSpringer-Verlag.Violations areliableforprosecutionundertheGermanCopyrightLaw. Springer-VerlagBerlinHeidelbergNewYork amemberofBertelsmannSpringerScience+BusinessMediaGmbH ©Springer-VerlagBerlinHeidelberg2000 PrintedinGermany Theuseofgeneraldescriptivenames,registerednames,trademarks,etc.inthispublication doesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexempt fromtherelevantprotectivelawsandregulationsandthereforefreeforgeneraluse. Typesetting:Camera-readybytheauthors/editors Coverdesign:design&production,Heidelberg Printedonacid-freepaper SPIN:10772918 55/3141/du-543210 Preface Modern physics is confronted with a large variety of complex spatial structures; almosteveryresearchgroupinphysicsisworkingwithspatialdata.Patternfor- mationinchemicalreactions,mesoscopicphasesofcomplexfluidssuchasliquid crystals or microemulsions, fluid structures on planar substrates (well-known as water droplets on a window glass), or the large-scale distribution of galax- ies in the universe are only a few prominent examples where spatial structures are relevant for the understanding of physical phenomena. Numerous research areas in physics are concerned with spatial data. For example, in high energy physics tracks in cloud chambers are analyzed, while in gamma ray astronomy observationalinformationisextractedfrompointpatternsofCherenkovphotons hitting a large scale detector field. A development of importance to physics in generalistheuseofimagingtechniquesinrealspace.Methodssuchasscanning microscopy and computer tomography produce images which enable detailed studies of spatial structures. Many research groups study non-linear dynamics in order to understand the time evolution of complex patterns. Moreover, computer simulations yield detailedspatialinformation,forinstance,incondensedmatterphysicsonconfig- urations of millions of particles. Spatial structures also derive from fracture and crack distributions in solids studied in solid state physics. Furthermore, many physicists and engineers study transport properties of disordered materials such as porous media. Because of the enormous amount of information in patterns, it is difficult to describe spatial structures through a finite number of parameters. However, statistical physicists need the compact description of spatial structures to find dynamicalequations,tocompareexperimentswiththeory,ortoclassifypatterns, for instance. Thus they should be interested in spatial statistics, which provides thetoolstodevelopandestimatestatisticallysuchcharacteristics.Nevertheless, until now, the use of the powerful methods provided by spatial statistics such as mathematical morphology and stereology have been restricted to medicine and biology. But since the volume of spatial information is growing fast also in physics and material science, physicists can only gain by using the techniques developed in spatial statistics. ThetraditionalapproachtoobtainstructureinformationinphysicsisFourier transformation and calculation of wave-vector dependent structure functions. Surely, as long as scattering techniques were the major experimental set-up in VI Preface ordertostudyspatialstructuresonamicroscopiclevel,thetwo-pointcorrelation function was exactly what one needed in order to compare experiment and the- ory.Nowadays,sincespatialinformationisevermoreaccessiblethroughdigitized images, the need for similarly powerful techniques in real space is obvious. In the recent decades spatial statistics has developed practically indepen- dentlyofphysicsasanewbranchinstatistics.Itisbasedonstochasticgeometry and the traditional field of statistics for stochastic processes. Statistical physics and spatial statistics have many methods and models in common which should facilitate an exchange of ideas and results. One may expect a close cooperation between the two branches of science as each could learn from the other. For in- stance,correlationfunctionsareusedfrequentlyinphysicswithvagueknowledge onlyofhowtoestimatethemstatisticallyandhowtocarryoutedgecorrections. On the other hand, spatial statistics uses Monte Carlo simulations and random fields as models in geology and biology, but without referring to the helpful and deep results already obtained during the long history of these models in statis- tical physics. Since their research problems are close and often even overlap, a fruitfulcollaborationbetweenphysicistsandstatisticiansshouldnotonlybepos- sible but also very valuable. Physicists typically define models, calculate their physical properties and characterize the corresponding spatial structures. But they also have to face the ‘inverse problem’ of finding an appropriate model for a given spatial structure measured by an experiment. For example, if in a given situation an Ising model is appropriate, then the interaction parameters need to be determined (or, in terms of statistics, ‘estimated’) from a given spatial con- figuration.Furthermore,thegoodness-of-fitoftheIsingmodelforthegivendata should be tested. Fortunately, these are standard problems of spatial statistics, for which adequate methods are available. Thegainfromanexchangebetweenphysicsandspatialstatisticsistwo-sided; spatial statistics is not only useful to physicists, it can also learn from physics. TheGibbsmodelsusedsoextensivelytodayinspatialstatisticshavetheirorigin inphysics;thusathoroughstudyofthephysicalliteraturecouldleadtoadeeper understanding of these models and their further development. Similarly, Monte Carlo simulation methods invented by physicists are now used to a large extent in statistics. There is a lot of experience held by physicists which statisticians should be aware of and exploit; otherwise they will find themselves step by step rediscovering the ideas of physicists. Unfortunately,contactbetweenphysicistsandstatisticiansisnotfreeofcon- flicts.Languageandnotationinbothfieldsareratherdifferent.Formanystatis- ticians it is frustrating to read a book on physics, and the same is true for statistical books read by physicists. Both sides speak about a strange language and notation in the other discipline. Even more problems arise from different traditions and different ways of thinking in these two scientific areas. A typical example, which is discussed in this volume, is the use of the term ‘stationary’ andthemeaningof‘stationary’modelsinspatialstatistics.Thiscanleadtoseri- ous misunderstandings. Furthermore, for statisticians it is often shocking to see how carelessly statistical concepts are used, and physicists cannot understand Preface VII theignoranceofstatisticiansonphysicalfactsandwell-knownresultsofphysical research. The workshop ‘Statistical Physics and Spatial Statistics’ took place at the University of Wuppertal between 22 and 24 February 1999 as a purely German event. The aim was simply to take a first step to overcome the above mentioned difficulties. Moreover, it tried to provide a forum for the exchange of fundamen- tal ideas between physicists and spatial statisticians, both working in a wide spectrum of science related to stochastic geometry. This volume comprises the majorityofthepaperspresentedorallyattheworkshopasplenarylectures,plus two further invited papers. Although the contributions presented in this volume areverydiverseandmethodicallydifferenttheyhaveonefeatureincommon:all of them present and use geometric concepts in order to study spatial configura- tions which are random. Toachievetheaimoftheworkshop,theinvitedtalksnotonlypresentrecent research results, but also tried to emphasize fundamental aspects which may be interesting for the researcher from the other side. Thus many talks focused on methodological approaches and fundamental results by means of a tutorial review.Basicdefinitionsandnotionswereexplainedanddiscussedtoclarifydif- ferentnotationsandtermsandthusovercomelanguagebarriersandunderstand different ways of thinking. Part 1 focuses on the statistical characterization of random spatial config- urations. Here mostly point patterns serve as examples for spatial structures. Generalprinciplesofspatialstatisticsareexplainedinthefirstpaperofthisvol- ume. Also the second paper ‘Stationary Models in Stochastic Geometry - Palm Distributions as Distributions of Typical Elements. An Approach Without Lim- its’ by Werner Nagel discusses key notions in the field of stochastic geometry and spatial statistics: stationarity (homogeneity) and Palm distributions. While a given spatial structure cannot be stationary, a stationary model is often ade- quateforthedescriptionofrealgeometricstructures.Stationarymodelsarevery useful,notleastbecausetheyallowtheapplicationofCampbell’stheorem(used as Monte Carlo integration in many physical applications) and other valuable tools. The Palm distribution is introduced in order to remove the ambiguous notion of a ‘randomly chosen’ or ‘typical’ object from an infinite system. In the two following contributions by Martin Kerscher and Karin Jacobs et al. spatial statistics is used to analyze data occurring in two prominent phys- ical systems: the distribution of galaxies in the universe and the distribution of holes in thin liquid films. In both cases a thorough statistical analysis not only reveals quantitative features of the spatial structure enabling comparisons of experiments with theory, but also enables conclusions to be drawn about the physical mechanisms and dynamical laws governing the spatial structure. In Part 2 geometric measures are introduced and applied to various exam- ples. These measures describe the morphology of random spatial configurations and thus are important for the physical properties of materials like complex fluids and porous media. Ideas from integral geometry such as mixed measures or Minkowski functionals are related to curvature integrals, which characterize VIII Preface connectivity as well as content and shape of spatial patterns. Since many phys- ical phenomena depend crucially on the geometry of spatial structures, integral geometry may provide useful tools to study such systems, in particular, in com- binationwiththeBooleanmodel.Thismodel,whichiswell-knowninstochastic geometryandspatialstatistics,generatesrandomstructuresthroughoverlapping random‘grains’(spheres,sticks)eachwithanarbitraryrandomlocationandori- entation. Wolfgang Weil focuses in his contribution on recent developments for inhomogeneousdistributionsofgrains.PhysicalapplicationsofMinkowskifunc- tionals are discussed in the paper by Klaus Mecke. They range from curvature energiesofbiologicalmembranestothephasebehavioroffluidsinporousmedia and the spectral density of the Laplace operator. An important application is the morphological characterization of spatial structures: Minkowski functionals leadtoorderparameters,todynamicalvariablesortostatisticalmethodswhich arevaluablealternativestosecond-ordercharacteristicssuchascorrelationfunc- tions. A main goal of stereology, a well-known method in statistical image anal- ysis and spatial statistics, is the estimation of size distributions of particles in patterns where only lower-dimensional intersections can be measured. Joachim Ohser and Konrad Sandau discuss in their contribution to this volume the es- timation of the diameter distribution of spherical objects which are observed in a planar or thin section. Ru¨diger Hilfer describes ideas of modeling porous media and their statistical analysis. In addition to traditional characteristics of spatial statistics, he also discusses characteristics related to percolation. The models include random packings of spheres and structures obtained by simu- lated annealing. The contribution of Helmut Hermann describes various models forstructuresresultingfromcrystalgrowth;hismaintoolistheBooleanmodel. Part 3 considers one of the most prominent physical phenomena of random spatialconfigurations,namelyphasetransitions.Geometricspatialpropertiesof asystem,forinstance,theexistenceofinfiniteconnectedclusters,areintimately related to physical phenomena and phase transitions as shown by Hans-Otto Georgiiinhiscontribution‘PhaseTransitionandPercolationinGibbsianParti- cle Models’. Gibbsian distributions of hard particles such as spheres or discs are often used to model configurations in spatial statistics and statistical physics. Suspensions of sterically-stabilized colloids represent excellent physical realiza- tionsofthehardspheremodelexhibitingfreezingasanentropicallydrivenphase transition.HartmutLo¨wengivesinhiscontribution‘FunwithHardSpheres’an overview on these problems, focusing on thermostatistical properties. In many physical applications one is not interested in equilibrium configu- rations of Gibbsian hard particles but in an ordered packing of finite size. The questionofwhetherthedensestpackingofidenticalcoinsonatable(orofballsin space)iseitherasphericalclusterorasausage-likestringmayhavefar-reaching physical consequences. The general mathematical theory of finite packings pre- sented by Jo¨rg M. Wills in his contribution ‘Finite Packings and Parametric Density’ to this volume may lead to answers by means of a ‘parametric density’ Preface IX which allows, for instance, a description of crystal growth and possible crystal shapes. Thelastthreecontributionsfocusonrecentdevelopmentsofsimulationtech- niquesattheinterfaceofspatialstatisticsandstatisticalphysics.Themainrea- son for performing simulations of spatial systems is to obtain insight into the physical behaviour of systems which cannot be treated analytically. For exam- ple, phase transitions in hard sphere systems were first discovered by Monte Carlo simulations before a considerable amount of rigorous analytical work was performed (see the papers by H. Lo¨wen and H.-O. Georgii). But also statisti- cians extensively use simulation methods, in particular MCMC (Markov Chain Monte Carlo), which has been one of the most lively fields of statistics in the last decade of 20th century. The standard simulation algorithms in statistical physics are molecular dynamics and Monte Carlo simulations, in particular the Metropolisalgorithm,whereaMarkovchainstartsinsomeinitialstateandthen ‘converges’ towards an equilibrium state which has to be investigated statisti- cally.Unfortunately,whetherornotsuchanequilibriumconfigurationisreached after some simulation time cannot be decided rigorously in most of the simu- lations. But Elke Tho¨nnes presents in her contribution ‘A Primer on Perfect Simulation’ a technique which ensures sampling from the equilibrium configu- ration, for instance, of the Ising model or the continuum Widomn-Rowlinson model. Monte Carlo simulation with a fixed number of objects is an important tool in the study of hard-sphere systems. However, in many cases grand canonical simulationswithfluctuatingparticlenumbersareneeded,butaregenerallycon- sidered impossible for hard-particle systems at high densities. A novel method called ‘simulated tempering’ is presented by Gunter Do¨ge as an efficient alter- native to Metropolis algorithms for hard core systems. Its efficiency makes even grand canonical simulations feasible. Further applications of the simulated tem- pering technique may help to overcome the difficulties of simulating the phase transition in hard-disk systems discussed in the contribution by H. Lo¨wen. The Metropolis algorithm and molecular dynamics consider each element (particle or grain) separately. If the number of elements is large, handling of them and detecting neighbourhood relations becomes a problem which is ap- proached by Jean-Albert Ferrez, Thomas M. Liebling, and Didier Mu¨ller. These authors describe a dynamic Delaunay triangulation of the spatial configurations based on the Laguerre complex (which is a generalization of the well-known Voronoi tessellation). Their method reduces the computational cost associated withtheimplementationofthephysicallawsgoverningtheinteractionsbetween the particles. An important application of this geometric technique is the simu- lation of granular media such as the flow of grains in an hourglass or the impact of a rock on an embankment. Such geometry-based methods offer the potential ofperforminglargerandlongersimulations.However,duetotheincreasedcom- plexity of the applied concepts and resulting algorithms, they require a tight collaboration between statistical physicists and mathematicians. X Preface It is a pleasure to thank all participants of the workshop for their valuable contributions, their openness to share their experience and knowledge, and for the numerous discussions which made the workshop so lively and fruitful. The editors are also grateful to all authors of this volume for their additional work; the authors from the physical world were so kind to give their references in the extended system used in the mathematical literature. The organizers also thank the‘Ministeriumfu¨rSchuleundWeiterbildung,WissenschaftundForschungdes LandesNordrhein-Westfalen’forthefinancialsupportwhichmadeitpossibleto invite undergraduate and PhD students to participate. Wuppertal Klaus Mecke Freiberg Dietrich Stoyan June 2000