The Materials Genome Shortening the Iteration Cycle in Materials Development First IBM PC invented Lithium ion Teflon S. Whittingham Velcro Titanium Polycarbonate Diamond-like Thin Films Human Genome Mapped Sony Gerbrand Ceder Massachusetts Institute of Technology GaAs Amorphous soft magnets MGI Meeting Georgia Tech, June 5 2014 At all levels in the materials development process there is significant iteration Predictive modeling and quantification can reduce time Iterations are too long and there are too many iterations ! Modeling may reduce iterations and shorten them Knowledge, Methods and Information required to make rational materials design happen Basic Science KNOWLEDGE INFORMATION Data/Facts METHODS Tools … but we know nothing There are about 50,000 to 200,000 known inorganic compounds Elastic constants: about 200 compounds Super conductors ≈ 1000 Dielectric constant ≈ 300-400 For almost every property we are below 1% in coverage …. Imagine … That one knows for every known compound • The bandgap, the effective masses, the Seebeck coefficient • The elastic constants, the dielectric constants, the absorption spectra • The point defect energies, solubilities, electronic signature of defects, mobility of defects and impurities, The stacking fault energy, the Peierls dislocation barrier • Surface energies, in common environments (aqueous, air, …) That you had the capability to query these properties for novel compounds upon request … We know nothing about most materials … Example: LiFePO4 • First synthesized in 1977 • Only identified as useful battery cathode in 1997 Example: MgB2 • First synthesized in 1953 • Only identified as high temperature superconductor in 2001 More quantitative materials selection and higher scale modeling Materials selection Inform higher scale models Automated Quantum Chemistry Calculations for screening Molecule database Redox Potential Solubility Stability Chemical motif design Reactivity Isotropic elastic properties Synthesis Testing Feedback Anisotropic elastic properties Work by Edwin Garcia (Purdue) Origin of the Materials Genome idea Compute the basic properties ( “genes” ) of materials across the world of compounds and disseminate that information to the materials community to enable rapid materials searching and design. 2005-2006: several company funded discovery projects (Duarcell, Umicore, Bosch, Samsung The Materials Project: Vision www.materialsproject.org Compute the basic properties ( “genes” ) of materials across the world of compounds and disseminate that information to the materials community to enable rapid materials searching and design. 2005-2006: several company funded discovery projects (Duarcell, Umicore, Bosch, Samsung Kristin Persson Anubhav Jain Daniel Gunter Maciej Haranszyk Wei Chen Gerbrand Ceder Shyue Ping Ong Dane Morgan David Skinner Shreyas Cholia Jack Deslippe Stefano Curtarolo Alan Dozier Raphael Fink Obtain basic compound properties through scalable ab-initio computations + + Ne Ne N N 1 e e 1 2 H = ∑ ∇i + ∑Vnuclear (ri ) + ∑∑ 2 i j≠i rj − ri i=1 i=1 Thought provoking: NERSC flagship Hopper could crunch through ~20k structures in 1 day… ICSD Other experimental databases User submissions Input processing & transformations StructureNotationalLanguage (SNL) Web apps Analysis Workflow Manager Post-processing and error-checking Supercomputing Resources Materials API ICSD Other experimental databases User submissions Input processing & transformations pymatgen StructureNotationalLanguage (SNL) • Robust materials analysis Custodian Web apps • Self-healing error recovery Analysis Fireworks • Smart workflow management Workflow Manager Post-processing and error-checking Supercomputing Resources Materials API Today’s Status: ! Over 49,000 compounds" " ! Growing monthly" " ! Multiple property sets " ! Multiple tools" Automatic Import, Analysis and Dissemination Benchmarking of ~70 reaction energies with experiments + GGA/GGA+U mixing scheme 1638 measured compound enthalpies 49,000 computed compounds 20,000 low –T Phase Diagrams > 50,000 reaction enthalpies Kristin Persson <kapersson@lbl.gov> [Matgen] Pourbaix Diagrams on the Materials Project > 70,000 Pourbaix Diagrams Available 4 messages support@materialsproject.org <support@materialsproject.org> Reply-To: matgen@nersc.gov To: matgen@nersc.gov Wed, Oct 2, 2013 at 2:41 PM A few hundred experimental Gibbs free energies of aqueous ions combined with 49,000 calculated solid materials Calculated 3 Pourbaix Diagrams on the Materials Project Potential vs. SHE (V) 2H O ---> 2 CALC. MnO4 2 Mn+++ O +4H +4 e Today, we are excited to announce the release of the Pourbaix diagram MnO 1 2 -app. Pourbaix diagrams are solid-aqueous phase diagrams as a function of MnO4 2H O+2e ---> H +2 MnOOH pH, standard hydrogen potential and composition that can be used to OH 0 2 + 2 - - - 2 Mn ++ - HMnO2 Mn(OH)2 -1 -2 Mn -3 -2 1 of 4 0 2 4 8 6 pH at 25 °C 10 12 14 16 10/3/13 6:48 PM Several thousand bandstructures Lots of stuff in the works … • Elastic constants • Surfaces • Point Defects • Finite Temperature properties (phonon based) • Thermal expansion • Thermal Conductivity • Seebeck Coefficient • … • Design environment planned Data available through web interface or through API An open platform for accessing LOTS of Materials Project data Identifier (formula, id, etc) Preamble Property https://www.materialsproject.org/rest/v1/materials/Fe2O3/vasp/energy Request type API keys available at https://www.materialsproject.org/profile Data type (vasp, exp, etc.) Build it and They will Come ? www.materialsproject.org ! Launched online Oct 2011 ! Grants: 2012 DOE Center, JCESR, U Madison NSF Center, JCAP ! Users: > 6600 registered; 100-300 per day; 20+ publications last year from users ! Course-ware : UC San Diego, UC Davis, UC Irvine, U Michigan, John-Hopkins, Cornell, … ! Partner institutions: UC Berkeley, UC San Diego, MIT, Duke, U Wisconsin, U Kentucky, U Louvain Belgium, Cambridge UK, Caltech, … ! Companies: Toyota, Sony, Bosch, 3M, Honda, Samsung, LG Chem, Dow Chemicals, GE Global Research, Intermolecular, Applied Materials, Energizer, Advanced Materials, General Motors, Corning, DuPont, Nippon Steel, L’Oreal USA, Caterpillar, HP, Unilever, Lockheed Martin, Texas Instruments, Ford, Bose, SigmaAldrich, Siemens, Raytheon, Umicore, Seagate, … JCESR: Beyond Lithium Ion CROSSCUTING SCIENCE MATERIALS GENOME Multivalent Intercalation Chemical Transformation Non-Aqueous Redox Flow Systems Analysis and Translation Cell Design and Prototyping Commercial Deployment TECHNOECONOMIC MODELING EDL DISTINGUISHING TOOLS END-TO-END INTEGRATION 21 The Crosscutting ‘Electrolyte Genome’ Electrolyte Genome A design platform for novel electrolytes and redox-active molecules Computational structural/chemical/property data for 104-105 solvents, salts, and redox molecules; organized for interactive searching and design Prioritized target property list for each thrust Multivalent Intercalation Chemical Transformation Non-Aqueous Redox Flow Recent Collaboration with JCAP May contain trade secrets or commercial or financial information that is privileged or confidential and exempt from public disclosure. JCESR Sandbox: The Electrolyte Genome May contain trade secrets or commercial or financial information that is privileged or confidential and exempt from public disclosure. Other collaborations • Nanoporous materials center (Minnesota) • Functionals for accurate bandgaps (Jacobsen Denmark) • CO2 capture materials (Cambridge U) • Thermoelectrics (Caltech) • Mg alloy corrosion (Volkswagen) • Interfaces (Intermolecular) – negotiation in process • Advanced Battery Materials (Samsung) May contain trade secrets or commercial or financial information that is privileged or confidential and exempt from public disclosure. Some important factors May contain trade secrets or commercial or financial information that is privileged or confidential and exempt from public disclosure. Experimental data leverages computed data Experimental data not just for “checking” Combination of experiment and computed data is highly synergistic. The two leverage each other Experimental formation energies of used to correct some of the worst errors of DFT L. Wang et al. Physical Review B, 73, 195107 (2006). Using experimental enthalpy of formation – the Us are determined for every TM to make sure that the enthalpy of reactions between binaries are reproduced Our solution: Different approaches in different chemical domains. Connect through reference states Elemental metals GGA METALS reference reactions Binary Oxides OXIDES GGA + U A. Jain, Formation Enthalpies by Mixing GGA and GGA+U Calculations, Physical Review B, 84 (4), 045115 (2011). Stability in Aqueous Media? Cu+: http://www.ccp1.ac.uk/ newsletters/Current/Sprik/ CarParrinello.html + 20,000 low –T phase diagrams From first-principles ? ◦ Computationally demanding ~◦ 300 aqueous 2+, Na+, Mgion 2+, Al3+, K+, Li+, Be 2+,Cu2+, Ag+, OH-, Al3+, Ca2+, Fe Gibbs free energies and SO3- … Dissolution of solids in water: Combining experimental and computed energies reference reactions Element or simple compound Solid world GGA/GGA + U Dissolved ion Aqueous solutions Experimental free energies K. Persson et al, Physical Review B, 85, 235438 (2012). Kristin Persson <kapersson@lbl.gov> [Matgen] Pourbaix Diagrams on the Materials Project > 70,000 Pourbaix Diagrams Available 4 messages support@materialsproject.org <support@materialsproject.org> Reply-To: matgen@nersc.gov To: matgen@nersc.gov Wed, Oct 2, 2013 at 2:41 PM A few hundred experimental Gibbs free energies of aqueous ions combined with 49,000 calculated solid materials Calculated 3 Pourbaix Diagrams on the Materials Project Potential vs. SHE (V) 2H O ---> 2 CALC. MnO4 2 Mn+++ O +4H +4 e Today, we are excited to announce the release of the Pourbaix diagram MnO 1 2 -app. Pourbaix diagrams are solid-aqueous phase diagrams as a function of MnO4 2H O+2e ---> H +2 MnOOH pH, standard hydrogen potential and composition that can be used to OH 0 2 + 2 - - - 2 Mn ++ - HMnO2 Mn(OH)2 -1 -2 Mn -3 -2 1 of 4 0 2 4 8 6 pH at 25 °C 10 12 14 16 10/3/13 6:48 PM Software Management Philosophies Open-source ◦ More eyes => robustness ◦ Contributions from all over the world Benevolent dictators ◦ Unified vision ◦ Quality control Clear documentation ◦ Prevent code rot ◦ More users • Pymatgen • Custodian Test, test, test ◦ Continuous integration to ensure code is always working • Fireworks Python Materials Genomics is the core analysis code powering the Materials Project v2.7.0 v2.9.9 www.pymatgen.org stats • Steady increase over the past year • > 1000 views per month on average Major new features / functionality • Support for ABINIT 7.6.1 (ABINIT group/UCL) • Defects (Haranczyk/LBNL) • Qchem (JCESR) • Robust units handling (UCSD/UCL) • XRD pattern simulation (UCSD) Major new users / fans > 500 commits over the last year. Pymatgen coders work Mon-Wed. # of active contributors has more than doubled! Developers and users come from > 30 countries spread across all continents (except Antarctica!). MGI – Broadening its impact Recommendations: Data as Infrastructure • Data becomes part of infrastructure. • Is it time for a user facility to support data production and management for materials science and engineering ? (i.e. not just computing). Similar to the protein databank ? • Is the materials community convinced of this need ? • Requirements for data preservation and storage will come your way. Unless we act proactively we may not like what will be imposed on us Recommendations: Data and Code houses • Compute: Center for high-throughput computed data and curation. Compute all that can be computed. Make highthroughput routine. When a new property method becomes available, scale it ! Agencies (NSF, DOE, …) may have to provide more commodity computing (i.e. not exascale) • Experimental Data: need a place to dump it. Use DOI and minimal descriptors. Don’t try to “catalogue” • Software: For problems where the physics is well established, commodity code is needed. Can communities pick a few of these topics ? Even Google can not find most experimental data Science, Jan, 2014 Smaller particle size in LMCO/C leads to higher capacity than LMCO Recommendations: Learn best practice for MGI driven Materials Design by setting up field specific teams • Create teams of experimentalists/theory + MGI efforts to demonstrate how data can be used in real materials design. Do this in different fields • Feedback from design exercises to MGI effort is valuable. To design materials, field-specific expertise is needed • E.g. team on thermoelectrics, carbon capture, light weight alloys, etc. Recommendations: Education • Complexity of Materials Science requires computation and modeling. Yin and Yang • Traditional MS&E is not used to having lots of quantitative information • Continue to educate our young scientists and workforce. • Engage the educational institutions more in MGI. They will only do this if MGI is reasonably permanent Charting the Materials Genome is Possible • Within ten years most basic properties of inorganic compounds can be determined computationally • Materials researchers will have access to all this data. What you do with it will determine your competitive advantage • Researchers will be able to sit behind a terminal and request the computed properties of modified compounds Voltage$(V)$ 2.5 3 3.5 4 4.5 5 5.5 0 500 Eq.$Oxygen$Release$µO2, Charged$State$(eV) '1 1000 '1.5 1500 '2 2000 '2.5 2500 '3 3000 '3.5 Eq.$Oxygen$Release$Temp., Charged$State$(eV) 2 0 '0.5 3500 '4 4000 '4.5 '5 oxides phosphates silicates borates sulfates 4500 … towards a materials genome Let’s set pragmatic goals, not wish lists Thanks to Wei Chen, Geoffroy Hautier, Anubhav Jain, Shyue Ping Ong, Kristin Persson Kristin Persson Anubhav Jain Daniel Gunter Maciej Haranszyk Wei Chen Gerbrand Ceder Shyue Ping Ong Dane Morgan David Skinner Shreyas Cholia Jack Deslippe Stefano Curtarolo Alan Dozier Raphael Fink