Springer Theses Recognizing Outstanding Ph.D. Research For furthervolumes: http://www.springer.com/series/8790 Aims and Scope The series ‘‘Springer Theses’’ brings together a selection of the very best Ph.D. theses from around the world and across the physical sciences. Nominated and endorsed by two recognized specialists, each published volume has been selected for its scientific excellence and the high impact of its contents for the pertinent fieldofresearch.Forgreateraccessibilitytonon-specialists,thepublishedversions includeanextendedintroduction,aswellasaforewordbythestudent’ssupervisor explaining the special relevance of the work for the field. As a whole, the series will provide a valuable resource both for newcomers to the research fields described, and for other scientists seeking detailed background information on specialquestions.Finally,itprovidesanaccrediteddocumentationofthevaluable contributions made by today’s younger generation of scientists. Theses are accepted into the series by invited nomination only and must fulfill all of the following criteria • They must be written in good English. • The topic should fall within the confines of Chemistry, Physics and related interdisciplinary fields such as Materials, Nanoscience, Chemical Engineering, Complex Systems and Biophysics. • The work reported in the thesis must represent a significant scientific advance. • Ifthethesisincludespreviouslypublishedmaterial,permissiontoreproducethis must be gained from the respective copyright holder. • They must have been examined and passed during the 12 months prior to nomination. • Each thesis should include a foreword by the supervisor outlining the signifi- cance of its content. • The theses should have a clearly defined structure including an introduction accessible to scientists not expert in that particular field. David D. O’Regan Optimised Projections for the Ab Initio Simulation of Large and Strongly Correlated Systems Doctoral Thesis accepted by The University of Cambridge, UK 123 Author Supervisor Dr. DavidD.O’Regan Prof.MikeC. Payne CavendishLaboratory CavendishLaboratory TCMGroup TCMGroup Universityof Cambridge Universityof Cambridge JJ ThomsonAvenue JJ ThomsonAvenue Cambridge,CB3 0HE Cambridge,CB3 0HE UK UK e-mail: [email protected] e-mail: [email protected] ISSN 2190-5053 e-ISSN 2190-5061 ISBN 978-3-642-23237-4 e-ISBN978-3-642-23238-1 DOI 10.1007/978-3-642-23238-1 SpringerHeidelbergDordrechtLondonNewYork LibraryofCongressControlNumber:2011936135 (cid:2)Springer-VerlagBerlinHeidelberg2012 Thisworkissubjecttocopyright.Allrightsarereserved,whetherthewholeorpartofthematerialis concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broad- casting,reproductiononmicrofilmorinanyotherway,andstorageindatabanks.Duplicationofthis publicationorpartsthereofispermittedonlyundertheprovisionsoftheGermanCopyrightLawof September 9, 1965, in its current version, and permission for use must always be obtained from Springer.ViolationsareliabletoprosecutionundertheGermanCopyrightLaw. Theuseofgeneraldescriptivenames,registerednames,trademarks,etc.inthispublicationdoesnot imply, even in the absence of a specific statement, that such names are exempt from the relevant protectivelawsandregulationsandthereforefreeforgeneraluse. Coverdesign:eStudioCalamar,Berlin/Figueres Printedonacid-freepaper SpringerispartofSpringerScience+BusinessMedia(www.springer.com) To my beloved parents Supervisor’s Foreword Density functional theory is remarkable. By searching over the single particle electron density alone, in principle, it provides the exact quantum mechanical ground state energy of a given system and the corresponding exact ground state electron density. It achieves this incredible feat via the exact density functional, which, for any input electron density, outputs the sum of the kinetic, Hartree and exchange-correlation energies for the ground state many-body wavefunction which generates this input electron density. As we add the final electrons to a semiconductor when filling the valence bands, the density functional tells us that the change in energy on adding each successive electron is almost constant. However, when we add an extra electron, which has to enter the conduction band, the exact density functional tells us that the change in energy must differ, discontinuously, from its previous value. The difference in these energies is, in fact, the band gap of the semiconductor and exact density functional theory reproduces its value precisely even though the single particle electronic density changes by an infinitesimally small amount each time an electron is added to the system. Additionally, although the contribution to the single particle electron density from a given set of orbitals might equate to four and a half electrons, for example, the exact density functional captures the knowledge that electrons cannot be divided and that this occupancy can only occur because the true many- body wavefunction is a superposition of configurations, perhaps one which puts four electrons in this set of orbitals while another represents five electrons. In situations like this, the exact density functional is capable of precisely deter- mining the energies associated with this many-body wavefunction. Unfortunately, we are not as clever as the true density functional. Our crude approximations of the density functional break some of the physical constraints on the true many-body wavefunction, with a consequent detrimental effect on the predicted energies and densities. For example, all of the available density func- tionals get band gaps wrong, often badly underestimating them, and numerous approximate functionals quite happily place non-integer numbers of electrons on particular sites. It is my view that no density functional that we may create will ever overcome all of these problems simultaneously, although it is often possible vii viii Supervisor’sForeword to overcome one shortcoming using an approximate functional designed for that purpose. However, while such a functional will, by construction, give better results for the targeted property, it may then give worse predictions of other properties than standard functionals and it is thus in no real way closer to the exact density functional. Given the complexity of many-body wavefunctions it is, perhaps, remarkable that available density functionals work as well as they do—often predicting physical properties to within an accuracy of a few percent. Furthermore, density functional theory allows us to perform predictive calculations on systems con- taining many thousands of atoms, while we can only compute many-body wave- functionsforahandfulofelectrons.Oneapproachtoalleviatetheshortcomingsof the available density functionals is simply to insert the physics that is missing in these approximate functionals. For instance, it has now become very common to add an explicit van der Waals interaction between the atoms as an additional contributiontothetotalenergyinadensityfunctionaltheorycalculation.Another widely used approach is the so called DFT ? U method whereby a Hubbard U interaction is added to reproduce the physics of strongly correlated localised electronic orbitals. The weakness of previous implementations of DFT ? U, in which the occupancy of the orbitals is constrained to be an integer number of electrons, was that the results depended on the choice of the projectors used to determinetheoccupancyofthelocalisedorbitals.Thisthesispresentsamethodin which these projectors may be determined self-consistently during the DFT ? U calculation, thus providing an approach to overcome this weakness in previous implementations. This approach has been implemented in the linear scalingdensityfunctionalcode ONETEP andisshowntoretainthelinearscaling of computational cost with system size. This thesis contains applications of this technique to bulk nickel oxide, ligated iron porphyrins of biological interest and the copper phthalocyanine dimer, as well as scaling tests on nickel oxide nano- clusters containing over 7,000 atoms. InordertodeveloptheprojectorselfconsistentDFT ? Umethodology,itwas necessarytomasterthefullmathematicalcomplexitiesoftensorialcalculusinthe context of electronic structure calculations. This thesis contains a detailed exposition on the use of nonorthogonal orbitals, the construction of contracted tensorial invariants, energy minimisation algorithms on curved spaces and the Christoffel symbol corrections needed to ensure that the density matrix retains its idempotency, to first order, as the functions in which it is expanded are updated. This thesis provides a very detailed, yet readable, account of these issues and could become the standard reference on this topic for the electronic structure community. Many technological materials rely on strongly correlated electronic systems for their functional properties and atoms that host strongly correlated electronic orbitalsarefound intheactivesites ofmanyproteins.DFT methods have usually struggledtodescribe such systemsaccurately andthe resultsof DFT ? Ustudies have fundamentally depended on the set of projectors used in such calculations. As a result of the work presented in this thesis, we are moved a step closer to the Supervisor’sForeword ix accurate and routine description of such systems using first principles quantum mechanical approaches. Cambridge, June 2011 Prof. Mike C. Payne Acknowledgments Thisdissertationcomes,ostensibly,asaculminationofoneman’slaboursoverthe past three years. That it is but, as Donne wrote, ‘‘No man is an island, entire of itself’’ and it is a pleasure take some time here to thank those organisations and individuals who have contributed to this work and to my life over this period. My research has been generously supported by the UK Engineering and Physical Sciences Research Council and the National University of Ireland. The Cambridge HPCS and, via the UK Car-Parrinello Consortium, the UK National SupercomputingServiceHECToRhaveprovidedmuchoftherequiredcomputing resources. Pembroke College has provided travel grants, much pastoral support and a welcoming home for a good part of my time in Cambridge. My sincere thanks extends to these organisations for their assistance. The design of this dissertation is derived from a style package due to Thomas Fink and Robert Farr, though any inconsistencies in the layout are purely of my own making. The Thomas Young Centre at Imperial College London has allowed me to make frequent visits to the Mostofi group, which has been my academic home away from Cambridge for the past three years. I would like to warmly thank my friends at Imperial for their excellent welcome and all that they have taught me. Onecouldnotaskforafriendlierandmorestimulatingworkenvironmentthan the TCM group at Cavendish Laboratory; I have very much landed on my feet in thatsense.ItwasagreatprivilegetoshareanofficewithJamieBlundellandJohn ‘‘Maestro’’Bigginsforthreeyears;Ithinkthatthesupportsharedtheremorethan outweighed the ample distractions! All members of TCM have enriched my experience in some way, but I would like to particularly acknowledge Andrew Morris, Hatem Helal, Alex Silver, Robert Lee, Jonathan Edge, Gareth Griffiths, Priyanka Seth, Danny Cole, Patricia Silas, Mark Robinson, Sˆian Joyce, Professor MarkWarnerandProfessorDavidKhmelnitskiifortheirsupportandadvice.Iam also grateful to the regulars at the TCM DFT meetings for their help and enthu- siasm.Last,butbynomeansleast,IextendmysincerestthankstoTraceyIngham and Michael Rutter for unstinting generosity with their time and expertise. xi xii Acknowledgments Much of my efforts have centred around the ONETEP code and it is my pleasuretoacknowledgeallofthedevelopersandfellowcontributorstothisgreat work for their patience and professionalism. In particular, I thank Simon Dubois, Peter Haynes and Chris-Kriton Skylaris, who also kindly proof-read Chaps. 5 and 6, for stimulating discussions and suggestions. Nicholas Hine deserves a very special mention and thanks for a great deal of time spent guiding me; I have learned a huge amount from him in many matters and it is doubtful whether linear-scaling could be achieved for DFT ? U without his help. ProfessorCharlesFalco,ProfessorStephenFahyandDr.MichelVandyckhave been my academic mentors prior to postgraduate study, and without their invaluable encouragement I might not have commenced this work at all. Dr. Jonathan Yates and Professor Matteo Cococcioni have helped to direct my researchviastimulatingdiscussions.Myexaminers,ProfessorNicolaMarzariand Professor EmilioArtacho, offered someveryhelpfuladvice andcommentsonthe manuscript. I am very grateful. The friends I have made in Cambridge have got me through this process and madeitpossible,we’ve shared manyupsanddowns.IparticularlymentionKatia Shutova, Michelle Rigozzi, Kelsey Edwardsen, Matt Smith, Taylor Hathaway- Zepeda,ElizabethDearnley,EmmaFirestone,MatthiasWivel,RobinPayne,Peter EvanandKrishnaaMahbubanifortakingcareofme,Icannotthankthemenough. I fear that my exile might be made permanent if I neglected to thank Linda Mason, Jennifer Lavin, Niall Johansson, Aoife FitzGibbon O’Riordan, David O’Farrell, Sinéad Rose and David Sheehan for their loyal friendship. I promise I will try harder to stay in touch. Thank you too to all at Munster Vintage Motor Cycle and Car Club. TheproximityofmydearfriendShaneMansfieldhasbeenaverygreatcomfort to me. Thánamair abhus anso le chéile sa bhád agus, le cúnamh Dé, is sa chaoi chéanna go bhfillfimíd thar n-ais aríst lá éigin. My teacher, advisor, critic, counsellor and friend; the game would be lost completelyifitwerenotfortheunwaveringguidanceandgenerosityofDr.Arash Mostofi. My obligation to Arash is great, I thank him a thousand times. I thank my supervisor, Professor Mike Payne, for his excellent advice over these years, long hours spent proof-reading and straight answers when I needed themmost.MikehasbeenanenthusiasticadvocateatimportantmomentsandI’m very grateful indeed. The love and kindness shown by Florence Paul over these years has truly kept me going. I hope to be repaying it for a very long time. Finally,Iwouldliketothankallofmyfamilyfortheirloveandmindingsince dayone.My newgoddaughter,Isabel, has brightenedupa verycold winterspent in writing. I would be lost without my wonderful sister, Aoife, and my beloved Mother and Father, Bernice and John, to whom all of this is dedicated.
Description: