• Machine-learning models of matter beyond interatomic potentials

    The combination of electronic structure calculations and machine learning (ML) techniques has become commonplace in atomistic modelling—ML interatomic potentials, for example, can now describe the potential energy surface of a material across many phases, including a wide range of defects. Looking ahead, however, it is the calculation of ML models that can predict properties beyond the interactions between atoms that might eventually allow integrated machine learning models to replace costly electronic structure calculations entirely. In the paper Learning the electronic density of states in condensed matter, recently published in Physical Review B, researchers led by Michele Ceriotti, head of EPFL’s Laboratory of Computational Science and Modelling, have taken a step in that direction with a new ML framework for predicting the electronic density of states (DOS). The technique has already been applied to understand electronic transitions in dense amorphous silicon, in a paper that came out in Nature today. 

  • MARVEL and CSCS: a partnership built for exa-scale computing in materials science

    The NCCR MARVEL has ‘Materials’ revolution’ in its name for a reason: it seeks to radically transform and accelerate the process of materials discovery and design. It seeks to do this by taking a computational approach, built on a platform of database-driven, high-throughput quantum simulations: if we could compute the properties of every material using electronic structure calculations, and if we could use that information to screen big databases for materials with just the right characteristics for a given purpose — then that would put us at an enormous advantage in the search for new and better materials, avoiding much of the trial and error that comes with experiments in the lab. Such a project requires robust computer resources. For one, the sort of calculations that MARVEL research relies on take a lot of computing time and data storage; as those calculations get ever more sophisticated, demands on computing power will only grow. For another, to turn the MARVEL project into a true revolution, computer resources must be organised in such a way that they can serve as a platform for the broader materials science community.

  • Novel high-conductivity 2D semiconductors identified in new research

    Using state-of-the-art density-functional perturbation theory and the Boltzmann transport equation, researchers led by Prof. Nicola Marzari, head of THEOS and NCCR MARVEL, investigated monolayer materials with outstanding transport properties to identify several high-conductivity materials. While some have only recently been discussed in the literature, others have never been presented in this context. Comparing the 11 monolayers in detail allowed the researchers to investigate how the strength and angular dependency of electron-phonon scattering drives key differences in transport performance. It also allowed them to show the limitations of selecting potentially interesting materials based on band properties alone.

  • New computational analysis introduces surface coverage as a descriptor for screening semiconductor catalysts for water splitting

    In the paper “Evaluation of Photocatalysts for Water Splitting through Combined Analysis of Surface Coverage and Energy-Level Alignment,” a team of MARVEL researchers led by Alfredo Pasquarello, head of EPFL’s Chair of Atomic Scale Simulation have used computational analysis to identify a promising catalyst candidate for potential use in water splitting.  

  • “Amon”-based ML approach excels in modelling quantum properties of a wide range of systems

    Anatole von Lilienfeld, professor at the Institute of Physical Chemistry at the University of Basel and project leader of Incubator Project 2 at NCCR MARVEL, and colleague Bing Huang have developed  transferable quantum machine learning models that combine atom-in-molecule based fragments, dubbed “amons," with active learning to overcome challenges currently preventing the widespread application of first-principles-based exploration of chemical space. In the paper "Quantum machine learning using atom-in-molecule-based fragments selected on the fly," they demonstrate the efficiency, accuracy, scalability, and transferability of the models for important molecular quantum properties, such as energies, forces, atomic charges NMR shifts, polarizabilities, and in systems ranging from organic molecules to 2D materials and water clusters to Watson-Crick DNA base-pairs. The article was recently published in Nature Chemistry.

  • Systematic approach quantifies chemical diversity of different MOF libraries

    Researchers at EPFL including MARVEL Deputy Director Berend Smit and colleagues at MIT have introduced a systematic approach to quantifying the chemical diversity of different metal-organic framework material libraries and then using these insights to remove certain biases. Though their works is focused on MOFs because there has been exponential growth in the number of studied materials, the question of how to correctly sample material design space is relevant to many classes of materials.

  • Atomistic modelling probes the behavior of matter at the center of Jupiter

    Researcher Michele Ceriotti, professor at the department of Materials Science at EPFL and project leader of MARVEL’s Design & Discovery Project 1, and colleagues in Cambridge and Zürich have developed a physics-based machine learning approach to examine the behavior of hydrogen at extremely high pressures. The model reveals evidence of continuous metallization, and so has significant implications for planetary science. More fundamentally, it shows the way ahead for a simulation-driven change of the way we understand the behavior of matter in fields as diverse as drug development and alloys for automobiles. The paper has just been published in Nature. 

  • Materials Cloud, AiiDA, cornerstones of MARVEL open science strategy, feature in Scientific Data

    The latest issue of Nature group’s Scientific Data journal features papers on the Materials Cloud, an Open Science Platform designed to enable the seamless sharing of resources in computational materials science as well as AiiDA, an open-source Python infrastructure that helps researchers automate and share computational workflows. Publication is a testimony to the ever-increasing adoption of the two tools that emerged from EPFL Professor Nicola Marzari’s Theory of Simulation and Materials (THEOS) group, and now the cornerstones of NCCR MARVEL’s Open Science strategy. 

  • Non-Abelian topological charge explains why certain Weyl points with different chirality do not annihilate upon collision

    Common wisdom holds that two Weyl points with different chirality will annihilate when they collide with each other. Now, combining theoretical arguments with first-principles calculations, researchers including Oleg Yazyev, Chair of Computational Condensed Matter physics at EPFL and scientist  QuanSheng Wu have predicted, on the contrary, that this is not the case in Weyl points (WP) occurring near the Fermi level of zirconium telluride. In the paper Non-Abelian reciprocal braiding of Weyl points and its manifestation in ZrTe, recently published in Nature Physics, they report their finding that a pair of Weyl points in the C2T invariant plane around the K point carry non-trivial values of the non-Abelian charge, the Euler number, and will not annihilate, but rather convert into a nodal line (NL) in the mirror invariant plane when applying uni-axial strain along the [001] direction.

  • MARVEL research highlighted in JCP special issue on machine learning in chemical physics

    Edited by NCCR MARVEL group leaders Michele Ceriotti and Anatole von Lilienfeld as well as colleague Cecilia Clementi of Rice University, the JCP Special Topic on “Machine Learning Meets Chemical Physics” features papers from the labs of Clemence Corminboeuf and Ceriotti at EPFL and von Lilienfeld at the University of Basel. The issue focuses on research that puts an emphasis on the interplay between machine learning and chemical physics, whether by incorporating physical principles and chemical intuition into model construction or by using machine learning to recognize new laws or general design principles. In the article below, we look at highlights of the research contributed from the three MARVEL labs. 

  • 2-D materials for ultrascaled field-effect transistors: One hundred candidates under the ab initio microscope

    Researchers led by Mathieu Luisier from the Institute for Integrated Systems (IIS) at ETH Zurich and NCCR MARVEL's Director Nicola Marzari at EPFL have set out to develop a comprehensive atlas of 2-D materials that might be capable of challenging currently  manufactured silicon-based transistors, so-called Si FinFETs. The new simulations are based on earlier results from Marzari and his team, who used complex simulations on the Swiss National Supercomuting Centre's supercomputer "Piz Daint" to sift through a pool of more than 100,000 materials to identify 1,825 from which it might be possible to obtain 2-D layers of material. The paper 2‑D Materials for Ultrascaled Field-Effect Transistors: One Hundred Candidates under the Ab Initio Microscope, recently published in the journal ACS Nano, identifies 13 particularly promising candidates. 

  • Automated Wannierisation for high-throughput computational materials design

    High-throughput computational materials design is an emerging field that looks set to accelerate reliable, cost-effective design and optimisation of new materials that feature specific desirable properties. Maximally-localized Wannier functions (MLWFs)—a means of representing the Bloch eigenstates of a periodic system—are tools that have been regularly used to compute certain advanced materials properties from first principles. Bringing the two approaches together has been complicated by the fact that generating MLWFs automatically and robustly without user intervention and for arbitrary materials is difficult. Researchers at Nicola Marzari’s THEOS lab and colleagues have now addressed this problem by proposing a procedure for automatically generating MLWFs for use in high throughput frameworks. In the interest of Open Science, they have also developed a virtual machine that allows researchers to perform their own simulations, either with different parameters or on new materials using this new protocol. The research was a collaboration between the EU H2020 E-CAM and MaX Centres of Excellence. A Success Story on the project can be found here.