Frontiers of first-principles simulations: materials design and discovery
- Nicola Marzari (Swiss Federal Institute of Technology, Lausanne, Switzerland)
- Gerbrand Ceder (Massachusetts Institute of Technology, Dept. of Materials Science and Engineering, Cambridge, MA 02139, USA)
- Matthias Scheffler (Fritz Haber Institute of the Max Planck Society (FHI), Berlin, Germany)
The urgency of accelerating scientific discovery, invention, and technology transfer is driven by the urgent needs of our society, where the footprint on the environment and the drain on non-renewable resources are posing some of the greatest risks to our future. Novel materials or materials with novel properties are often the breakthroughs needed for such key societal advances – from the iron catalyst for ammonia synthesis, dramatically changing food production, to doped silicon empowering the information-and-communication technologies (ICT) revolution, to lithium-ion battery cathodes for portable devices, to giant-magnetoresistors for data storage. We need new materials to harvest solar energy, store it into fuels or into batteries, distribute it efficiently over the power grid, harness waste heat, efficiently convert electricity into mechanical energy, and sustain our information-and-communication revolution, to cite only a few examples.
Traditional materials development relies on intuition, intelligence, and extensive trial-and-error in a laboratory, with timelines that are often not compatible with the urgency of the broad societal needs alluded to above. Thanks to computer simulations, we have now some of the most powerful and novel tools ever available for scientific discovery and technological advancement – from understanding structure, stability, and performance at the nanoscale, to rapidly screening and testing novel generations of materials for strategic technological applications. This discipline has seen an exponential explosion in its relevance and impact, thanks to the combination of novel theoretical ideas, novel algorithms, and the ever increasing and inexpensive computational power sitting squarely on top of Moore’s law. In particular:
- Sheer computational power has been on a steady trend where it doubles every 14 months (J. Dongarra, Conference on Computational Physics, Nov 2011, considering the average of the top-500 most powerful computers in the world during the past 20 years [Dongarra2011]); this is compounded by a simultaneous improvement of the algorithms and the codes used in quantum simulations. In addition, high-throughput calculations are intrinsically massively parallel, and can be replicated in heterogeneous architectures, from commodity clusters to supercomputers, making them ideally suited for future computational applications (the hardware timeline has exaflop supercomputers entering operations in 2018).
- In the last few years quantum-mechanical simulations have become able to tackle complex systems with realistic, predictive accuracy - driven both by a novel generation of extensively validated functionals in density-functional theory, and by novel many-body techniques efficient enough to capture complex, correlated-electrons physics in realistic systems. The relevance of quantum simulations can be gauged, however cursorily, even by sheer bibliometric measures – in the list of the top 20 most cited papers in the history of the American Physical Society (i.e. from 1893 onwards) 18 papers, including the top 16, are on quantum simulations.
- Compositional and configurational space for new materials can be explored ever more efficiently thanks to the combination of novel sampling and structure-prediction methods [Ceder2006], neural-network potentials and machine-learning techniques, and more broadly by the combination of the concepts and algorithms of computer science as applied to the wealth of data produced by computational laboratories, and giving rise to the nascent field of materials’ informatics.
There is great strength in this combination. In a world where the traditional materials’ development cycle of “lab-to-fab” is on average ~20 years (think lithium-ion batteries, diamond-like carbon, amorphous soft magnets [Eager1995]), and the urgency of solving societal problems ever more pressing, we stand to greatly accelerate this process of invention and discovery by predictive, realistic, massively-parallel, database-driven and database-searching high-throughput simulations, at a time where solutions obtained through the traditional scientific pipeline might not arrive fast enough.
In fact, the field is ready to transition to a novel model for invention and discovery, where accurate, predictive, validated, and inter-operable quantum engines are put in the hands of a computer-science infrastructure that uses a novel generation of tools (massive heterogeneous databases, structure predictors, data mining and machine learning, and high-throughput calculations) to discover or optimize materials and properties [Curtarolo2013].
Hints of this revolution are right now emerging. In the US the White House has announced a “Materials Genome Initiative for Global Competitiveness” [MaterialsGenome2011], a “new, multistakeholder effort to develop an infrastructure to accelerate advanced materials discovery and deployment in the United States” with a “vision of how the development of advanced materials can be accelerated through advances in computational techniques, more effective use of standards, and enhanced data management”. [MG2011,MG2013]
In order to put such a plan into action, key components are missing – we need to create a materials informatics infrastructure able to run the quantum engines in a database-driven, database-filling mode [Jain2011], and we need to organize such databases to store heterogeneous information, while being able to data mine the information contained, and use machine learning and data analytics to uncover correlations between microscopic data and macroscopic performance. We need to establish rules and protocols for the verification of the computer codes, and the systematic validation against experiments of the calculated properties. We need to continue sustaining a robust effort in software engineering, implementation on distributed architectures, and in particular we need to keep pushing the scientific frontiers of electronic-structure simulations to deal with some of the current weak points in describing interacting electrons.
The CECAM and Psi-k communities have been at the forefront of novel developments in computational simulations, and we want to make sure this expertise is leveraged and all synergies explored to make sure that the entire community can benefit and focus of the current state-of-the-art in materials design and discover, and that it can target the most pressing areas for societal development. At the same time, we want to make sure we keep pushing the frontiers of electronic-structure simulations, of statistical physics and statistical mechanics, of large-scale and multi-scale simulations, and we integrate and complement this with state-of-the-art tools and concepts coming from the computer science community.
Ali Alavi (University of Cambridge, UK)
Jörg Behler (Ruhr-Universität Bochum, Germany)
Roberto Car (Princeton University, USA)
Gerbrand Ceder (Massachusetts Institute of Technology, Cambridge, USA)
Gábor Csányi (University of Cambridge, UK)
Michele Ceriotti (EPFL Lausanne, Switzerland)
Stefano Curtarolo (Duke University, Durham, USA)
Ralf Drautz (Ruhr-Universität Bochum, Germany)
Olle Eriksson (Uppsala University, Sweden)
Ralph Gebauer (International Center of Theoretical Physics, Trieste, Italy)
Luca Ghiringhelli (FHI, Berlin, Germany)
Stefan Goedecker (University of Basel, Switzerland)
Georg Gottlob* (University of Oxford, UK)
Karsten Jacobsen (Technical University of Denmark, Kongens Lyngby, Denmark)
Boris Kozinsky (Robert Bosch RTC, Cambridge, USA)
O. Anatole von Lilienfeld (University of Basel, Switzerland)
Volker Markl (TU Berlin, Germany)
Risto Nieminen (Aalto University, Finland)
Michele Parrinello (ETH, Zürich, Switzerland)
Kristin Persson (Lawrence Berkeley National Laboratory, USA)
Chris Pickard (University College London, UK)
Krishna Rajan (Iowa State University, Ames, USA)
Ramamurthy Ramprasad (University of Connecticut, USA)
Xinguo Ren (University of Science and Technology of China, Hefei, China)
Matthias Rupp (University of Basel, Switzerland)
David H. Vanderbilt (Rutgers, The State University of New Jersey, Piscataway, USA)
Chris Wolverton (Northwestern University, Evanston, USA)
*) to be confirmed
[Ceder2006] C. Fischer, K. Tibbetts, D. Morgan, G. Ceder, Predicting Crystal Structure: Merging Data Mining with Quantum Mechanics, Nature Materials, 5 (8), 641-6 (2006).
[Curtarolo2013] S. Curtarolo and G. L. W. Hart and M. B. Nardelli and N. Mingo and S. Sanvito and O. Levy, Nature Materials 12, 191 (2013).
[Jain2011] A. Jain, G. Hautier, C.J. Moore, S.P. Ong, C.C. Fischer, T. Mueller, K.A. Persson, G. Ceder, A High- Throughput Infrastructure for Density Functional Theory Calculations, Computational Materials Science, 50 (8) 2295-2310 (2011).