“BigData”inmaterials modelling AngelosMichaelides ThomasYoungCentre, LondonCentreforNanotechnology&DepartmentofPhysics& Astronomy, UniversityCollegeLondon(UCL),UK www.chem.ucl.ac.uk/ice www.thomasyoungcentre.org Whatismaterials modelling? - Alsoknownasthetheory andsimulationof materials - Increasinglyinvolvesthe applicationofcomputers tounderstandthe propertiesofexistingand newmaterials. Themolecularmodelingrevolutionin chemistry,physicsandbiology Kilopapers 20 15 TOTAL Force field Density functional theory Wave function theory 10 5 1991 1996 2001 Year 2006 2011 • Molecularlevelinsighttoexistingmaterials • Examplesofmaterialsdiscoveryto:Catalysts,Batteries, Highstrengthalloys,Hydrogenstorage,Thermoelectrics,… Keywords used: Topic=("HF" or "WFT" or "MP2" or "MCSCF" or "Hartree-Fock" or "post Hartree-Fock" or Water: Themostanomalousliquid …themostubiquitous,essentialforlife Contemporaryconcernsforsociety: (thatsciencecanhelpdirectlywith) ClimateHealthEnergy Directlyorindirectlyrelatedto water;generallyatinterfaces Materials modelling is now “data rich” - Increasedcomputationalcapacity,improvedalgorithms andcodesmeansmuchshortertimetosolution - Bigdataissues,e.g.: Identificationofnew structures,polymorphs ii) Exploration of processes previously beyond reach iii) More accurate evaluations of e.g. E (energy) HΨ = EΨ i) Nature Materials 15, 66 (2016) 1. New structures, polymorphs, materials - Variousinternationaldatabasesemerging,e.g.MaterialsProject (Berkeley/MIT), NoMaD (Berlin),CPOSS(UCL) - Confinedwater:Crystalstructurepredictions:10,000+trial structures,mining toidentifyinterestingones,descriptors to explainproperties… Phys.Rev.Lett116,025501(2016) 2.Newprocesses,e.g.iceformation - Freezingofwater:notunderstoodatthemolecularlevel - Mechanismdeterminedusingforwardfluxsampling– involving ~100,000trajectories(eachcontaining30,000atoms) Patternrecognition: • 5 million CPU hours • 50,000 configurations involved in the I/O • 5 Terabytes of data 3. More accurate energies, e.g. water Mostwidelyusedquantumapproach(DFT)forsimulations ofliquidwaternotuptothejob… Reference StandardDFT Prism 0.00 0.00 Cage 0.24 -0.59 Book 0.70 -2.18 Cyclic 1.69 -2.44 J.Chem.Phys.144,130901(2016) Gaussianapproximationpotential - Databaseofaccurate(~exact)energiesfor watermonomers,dimers,trimers - FitgeneralinteractionwithGaussianregression Reference StandardDFT Machine Learning Prism 0.00 0.00 0.00 Cage 0.24 -0.59 0.15 Book 0.70 -2.18 0.52 Cyclic 1.69 -2.44 1.77 - Verygoodperformanceforiceandliquidwater… TheThomasYoungCentre • Interdisciplinaryallianceofabout80groupsworking toaddresschallengesofsocietyandindustrythrough theory&simulationofmaterials • Imperial,King’s,QMUL,UCL;Chemistry,Physics, EarthSciences,Materials,Engineering,Nanotech. • Ahubforcollaborations&aportaltoastrong interdisciplinaryresearchcommunityoperatingatthe forefrontofscience;placing Londonatcentre ofinternational materialsmodelling • Formoreinformationsee: www.thomasyoungcentre.org orthismonth’sedition ofNatureMaterials(Volume15,page371(2016)) Conclusions • Bigdataisabigissueinmaterials modelling • Bigdatarelevanttounderstanding importantphysiochemicalproblems www.chem.ucl.ac.uk/ice • UCLmodellingandTYCarestrong inthisfield • Successreliesonextracting physicalinsightfromdataand accesstoworldclasscomputing facilitiestogenerateit… www.chem.ucl.ac.uk/ice www.thomasyoungcentre.org Free game on iTunes