Highlights

Using a new workflow based on density functional theory, perturbation theory and Wannier functions, a team including MARVEL researchers has scanned large databases of experimental materials to identify potential new superconductors to be used in applications such as Magnetic Resonance Imaging. The search resulted in 24 promising candidates with critical temperature above 20K, which would allow to cool them with liquid hydrogen instead of the more expensive liquid helium. The article is published in PRX Energy. 

A new study by MARVEL researchers at Empa and EPFL has explained a perplexing discrepancy between theoretical predictions and experimental data on copper-tungsten nanomultilayer, materials made by stacking alternating, ultra-thin layers of the two materials. Experiments show a transition from compressive to tensile interface stress when thickness and pressure pass a certain threshold, but DFT calculations could not model it. Thanks to a neural-network potential developed by Bill Curtin's group at EPFL, Vladyslav Turlo and his team at Empa showed that the transition is due to intermixing of the two elements at the interface, a result then validated by experiments. The study is published in Acta Materialia. 

A new study involving former MARVEL member Michele Simoncelli and  MARVEL director Nicola Marzari combines a first-principles approach with machine learning to identify a unique material with distinctive thermal properties. The material combines crystal and glass thermal properties and is the first of its kind. It was discovered in meteorites and has also been identified on Mars. The fundamental physics driving this behavior could advance  understanding and design of materials that manage heat under extreme temperature differences—and, more broadly, provide insight into the thermal history of planets. The study is published in PNAS.

In a new study just published in npj Computational Materials, Vladyslav Turlo and his group at Empa in Thun introduce a new, fast atomistic simulation technique that leverages spectroscopy data, machine learning, and molecular dynamics techniques to model how hydrogen interacts with thin alumina films manufactured through atomic layer deposition (ALD). By simulating how hydrogen is incorporated in these amorphous alumina structures, the study can help their application - for example for the inner barrier shell of a hydrogen storage container.  In the future, the technique could also help developing new membranes for hydrogen purification or gas separation. 

A new article by MARVEL scientists published in Physical Review B presents a new computational framework to simulate the behavior of tetragonal barium titanate (BTO), a ferroelectric perovskite that can be used as an alternative to silicon in photonic integrated circuits, that use photons instead of electrons to encode and transmit information. The core of the study is the modelling of the Pockels effect, a change in the refractive index of the material in the presence of an electric field that can be used to encode information. The authors, from Mathieu Luisier’s lab at ETH Zurich and Nicola Marzari’s group at EPFL, devised a computational framework that is independent of specific functionals and relies only on standard Density Functional Theory. The results were validated by comparing them to existing experimental results, and to previous calculations based on other computational methods.

The theoretical description of spin excitations that occur when atoms in a solid are perturbed by an external magnetic field remains a challenge for computational materials science. A new article in npj Computational Materials by MARVEL members Luca Binci, Iurii Timrov and Nicola Marzari introduces a new method to calculate magnetic excitations in transition-metal oxides fully ab initio, without relying on simplified mathematical models, or without making arbitrary empirical assumption. The method combines DFT with Hubbard functionals (DFT+U) and time-dependent DFT (TDDFT), and was validated on manganese oxide and nickel oxide, two prototypical transition-metal oxides. In the future it can be used to study exotic magnetic materials. 

During the first and second phases of MARVEL, theorists and experimentalists collaborated on identifying and testing promising topological materials, that owe their behaviour to the geometrical properties of their electronic structure. EPFL's Oleg Yazyev led the project, with key experiments performed by the group of Ming Shi, then at the Paul Scherrer Institute. It all began with the creation of a database of topological materials, and continued with groundbreaking studies on bismuth iodide, followed by the Weyl semi-metals tantalum phosphide (TaP) and tungsten phosphide (WP2). The insight gained from research on topological materials has spilled out on other areas of physics. This is the second article in a series about MARVEL's success stories from its 10 years of research. You can read the first one here. 

A collaboration between groups in Switzerland and Italy has proposed a fully automated workflow that makes it much easier for scientists to calculate the stopping site of muons and their interaction with the environment during muon spectroscopy, increasing the power of this experimental technique when it comes to studying fine details in the magnetic properties of materials.  The workflow integrates existing codes and libraries with newly developed algorithm and takes advantage of the AiiDA infrastructure. The article that describes it was published in Digital Discovery.

A study by MARVEL researchers presents a new framework to handle hybrid quantum-classical algorithms for materials simulations. The framework works by interfacing two widely used software tools that belong respectively to the classic and quantum world: CP2K and Quiskit Nature.  The key step is to identify the "active space" in a material - the orbitals and electrons that are key for the properties of interest. Quantum algorithms are used to simulate the active space at a higher level of detail, while classical ones are applied to the rest of the system.  When tested on magnesium oxide, the workflow produces results that are in very good agreement with those from state-of-the-art theoretical and experimental methods. The study is published in npj Computational Materials.

Scientists at EPFL and the Paul Scherrer Institute have shown that machine learning can reduce the time and computational cost of density-functional theory with extended Hubbard functionals, a widely used method that allows to simulate complex materials containing transition-metal or rare-earth elements. Using a recently developed class of neural networks called “equivariant neural networks”, and a dataset of 12 materials spanning various crystal structures and compositions, the team trained two separate models – one for the U parameter and one for the V – to work independently of one another. The models performed very well in calculating both the U and V parameters themselves, as well as some downstream properties such as magnetic moments or voltages. The study is published in npj Computational Materials.

A new review article, just published in Reviews of Modern Physics and highlighted on the journal cover, provides a map to the vast landscape of software codes that allow researchers to calculate Wannier functions, and to use them for materials properties predictions.  The authors, from all over Europe and the USA, include three current MARVEL members and three former ones. After providing readers with the theoretical foundations on Wannier functions and their calculation, together with intuitive graphical schematics to explain what Wannier functions are, the authors map the existing Wannier codes and the key applications. Several codes that now make up the Wannier ecosystem were developed within or with the support of MARVEL.

MARVEL scientists at the Paul Scherrer Institute and the University of Zurich have used a machine learning model to calculate the screening parameters for Koopmans functionals, a promising approach to expand the power of density-functional theory so that it can be used to predict the spectral properties of materials. The study, published in npj Computational Materials, focussed on two model systems: liquid water and the halide perovskite CsSnI3. Even with a relatively simple network and learning model, the scientists were able to significantly reduce the computational cost of the algorithm, paving the way to a more efficient calculation of the temperature-dependent spectral properties of interesting materials.