Improving the generation and sampling of crystal packings with machine learning
PI: Prof. Sereina Riniker, Dept. of Computational Chemistry at ETHZ
Integrated into Design & Discovery Project 1
Organic molecules can have different polymorphs (i.e. crystal packing motifs), resulting in varying physical properties such as solubility. Changes in solubility can in turn have a significant impact on the effects of a given drug compound, thus identifying possible polymorphs is essential. Sampling crystal packings by carrying out many crystallization experiments under different conditions can help researchers develop databases to assist with this identification, but the process is labor-intensive and difficult to scale up.
Using computational tools such as machine learning (ML) in crystal structure prediction helps in the search for stable solid forms. The ultimate aim of Improving the generation and sampling of crystal packings with machine learning is to develop general ML models that can assist in crystal structure prediction for organic compounds.
There are, in general, two main challenges to developing such models. The first is linked to the sampling of potential crystal packings, and the second to scoring or ranking the generated packings. Riniker's project focuses on the first challenge.
As a molecule becomes more flexible, the search space grows exponentially—both the possible conformations of a single molecule as well as the relative arrangements in the crystal have to be sampled. Figuring out how to reduce efficiently and accurately the number of structures to be scored is then essential. The basic idea behind the proposed approach is to shift from complete coverage of the configurational space to a more targeted subset of structures that are likely to be low in energy and therefore stable. This could speed up the crystal structure prediction process significantly and allow for the application of more cost-intensive methods earlier in the crystal-structure prediction process.
Using this approach, the researchers plan to learn molecular arrangements from available crystal structure databases using ML, to then generate candidate crystal packings with the trained models, and, finally, to pre-filter the generated structures.
"We are aiming to develop low-cost ML methods to produce high-quality candidate structures that can then be supplied to scoring and ranking by more accurate, but computationally more expensive, methods," Riniker said.
Search for MOF-based catalysts for the electrochemical splitting of water
PI: Dr. Emiliana Fabbri, Electrochemistry Laboratory, Paul Scherrer Institute
Integrated into Design & Discovery Project 4
Electrochemical water splitting is a critical area of research in the development of renewable energy because the majority of such sources vary in supply based on factors such as season and region. Introducing renewable energy sources on a wide scale then will only be possible with the development of reliable energy storage systems that can store large amounts of energy over the long term. Electrolyzers— electrochemical devices that can split water molecules into H2 and O2, thereby converting the excess of secondary energy from renewable sources into chemical energy "stored" into the hydrogen chemical bonds—are expected to play an essential role in the process.
The performance of water electrolyzers is mostly linked to the anodic reaction, that is, the oxygen evolution reaction (OER). The search for highly active anodic catalysts is ongoing and very active—promising candidates should show electrochemical activity towards the OER, but also stability under operating conditions. The development of cost-effective, robust and highly active anode materials would represent a breakthrough in development and introduce the potential for the widespread commercialization of water electrolyzers. Recently, research has shown that MOF-based Co or Ni catalysts are promising candidates for this use.
"The Agility Plus project aims at understanding and engineering the fundamental parameters governing the electrochemical water splitting reaction on the surface of MOF-based catalysts by means of a close synergy between theoretical calculations and experimental verifications," Fabbri said. "The development of efficient and durable catalysts based on abundant elements for water electrolyzers is a vital step toward the widespread implementation of renewable energy and, thus, zero-carbon emission energy sources."
Fabbri's project aims to systematically investigate the nature of the OER activity of MOF-based catalysts (i.e., active site, reaction mechanism, activity descriptors) and to screen different MOF-based catalysts in an effort to identify the most active one. Specifically, the project will synthesize and characterize a series of MOF-74(M) based catalysts (M = Mn, Fe, Co, Ni, Cu and Zn) and binary metal MOF catalysts, correlating physical and chemical properties such as surface area and composition, morphology, structure, ex situ electronic configuration their respective OER activities. They will then measure the catalysts with X-ray absorption spectroscopy and correlate these results with the OER activity. At the same time, they will predict the oxidation state for several MOF-based compounds using a theoretical machine learning approach developed by Berend Smit, head of the Laboratory of Molecular Simulation at EPFL and leader of project D&D4. Finally, the group will look to correlate the experimental and theoretical results.
Electronic properties of SrCr03 thin films
PI: Prof. Marta Gibert, Group Leader, Oxide Interface Physics, University of Zurich
Integrated into Design & Discovery Project 5
Complex transition metal oxides (TMOs) are a fascinating class of compounds that display a wide spectrum of physical properties that result from an intricate interplay between the charge, spin, orbital and lattice degrees of freedom. Competition between correlated electron phases leads to complex phase diagrams and strong sensitivity to external parameters such as temperature and magnetic field. What's more, we can now generate oxide thin films and heterostructures, opening the door to further tuning of those bulk properties, even allowing for the stabilization of metastable phases and/or the emergence of novel functionalities.
Cr-based TMOs are difficult to synthesize and have been little explored—the electronic properties of some of these compounds are controversial. For instance, depending on the samples, most of which are polycrystalline, SrCrO3 has been reported to display metallic or semiconductor behavior, paramagnetism or antiferromagnetism at low temperature. The most recent studies on 50 nanometer-thick SrCrO3 films indicate a correlated metallic behavior with potential signatures of antiferromagnetism at low temperatures. The coexistence of antiferromagnetism and metallicity is an interesting, uncommon phenomenon that is poorly understood.
In the project Electronic properties of SrCr03 thin films, Gibert and her team will investigate and clarify the electronic properties of SrCrO3 by growing high quality epitaxial thin films, focusing mainly on the transport and magnetic properties of the material as function of epitaxial strain and film thickness. Afterwards, they intend to generate more complex heterostructures, allowing them to experiment with interface phenomena such as charge transfer, giving them a rich playground for understanding and modulating the electronic properties of such strongly correlated compound. The outcome of the project is key for the realization of future devices, which could exploit the potential ferroelectricity, thermoelectricity or the metallic antiferromagnetic ground state of SrCrO3. The investigation is also of theoretical interest because the compound may prove to be a so-called Hund's metal, in which correlations are mainly caused by the Hund's rule coupling.
Topological materials with intrinsic magnetic ordering
PI: Prof. Ana Akrap, Department of Physics, University of Fribourg
Integrated into Design and Discovery Project 6
Incorporating magnetism is a key goal in the design of novel topological materials—together, magnetism and topology can lead to exotic phenomena such as anomalous quantum Hall effect, half-integer quantum Hall effect or the effective realization of axion electrodynamics.
In the project Topological materials with intrinsic magnetic ordering, researchers will experimentally investigate a family of compounds that promises to host topological effects within intrinsically magnetically ordered systems: MnBi2Te4, EuIn2As2, EuCd2As2, EuMnBi2 and YbMnBi2.
Magnetically ordered topological insulators were first created through doping with 3D-metals. This created inhomogeneity and reduced the performance of the resulting materials. Topological insulators with intrinsic magnetic ordering, such as that found in the MnBi2Te4 and the EuIn2As2 families, are therefore highly desirable—their low disorder is linked to better material performance.
For instance, MnBi2Te4 has a large gap in the topological surface states due to antiferromagnetic (AFM) ordering and may offer a realization of axion electrodynamics and half-integer quantum Hall effect. EuIn2As2 with AFM ordering is predicted to be a gapped topological insulator and could also lead to a possible realization of axion electrodynamics. The EuIn2As2 family of materials shows rich magnetic phase diagrams, where the type of magnetic ordering changes with the strength of magnetic field. YbMnBi2 is possibly an ideal Weyl semimetal, with claims of a pair of Weyl points located at the Fermi level. YbMnBi2 is an antiferromagnet, but due to possible canting of spins, an in-plane ferromagnetic component is created, leading to a possible Weyl semimetal phase.
The overall goal of the project is to unravel and better understand these varied properties through an approach that has already been shown to work well for topological systems. Using high-quality single crystals with large surfaces as samples, the group will perform optical and magneto-optical spectroscopy coupled with first principles band structure calculations and effective models. Optical spectroscopy is ideally suited to address the electronic structure at the milli-electron-volt scale, which is where the most interesting physics happens, while magneto-optical experiments in high fields allow Landau-level spectroscopy, a very powerful way of understanding the topological nature of the electronic bands.
Low-volume newsletters, targeted to the scientific and industrial communities.Subscribe to our newsletter