Add BrightSurf on Google Email

National Centre of Competence in Research (NCCR) MARVEL


New model makes machine learning potentials more accurate and more accessible

A new dataset and model improve the efficiency of machine-learning interatomic potentials and their applicability to different chemical elements and material classes. The PET-MAD model uses a compact and denser dataset of 95,595 structures and an original neural network architecture, achieving robust simulations with minimal fine-tuning.

How machine learning can help predict the spectral properties of materials

A new study uses machine learning to reduce time needed for calculating screening parameters in Koopmans functionals, enabling faster predictions of material spectral properties. Researchers trained a simple model using modest data and achieved accurate results, paving the way for studying temperature-dependent spectral properties.

SourceNational Centre of Competence in Research (NCCR) MARVEL·Journalnpj Computational Materials·TypeComputational simulation/modeling·DateDec 22, 2024

In search of the perfect materials for fusion reactors

Researchers used computational methods to screen potential plasma-facing materials for fusion reactors, considering factors like thermal resistance and neutron bombardment. A shortlist of 21 materials was identified, including tungsten, diamond, and tantalum nitride, which showed promise for divertor applications.

SourceNational Centre of Competence in Research (NCCR) MARVEL·JournalPRX Energy·TypeComputational simulation/modeling·DateNov 5, 2024

Computational marathon matches the efficiency of the AiiDA platform with the power of Switzerland Alps supercomputer

A team of scientists successfully interfaced AiiDA with the Alps supercomputer, completing over 100,000 calculations in just 16 hours. The run demonstrated the maturity of Swiss-made software tools for computational materials science and showcased the power of Switzerland's main supercomputing facility.

A “gold standard” for computational materials science codes

A team of scientists from NCCR MARVEL has carried out the most comprehensive verification effort on solid-state DFT codes, simulating 960 materials and their properties using two independent state-of-the-art codes. The study provides a reference dataset and guidelines for assessing and improving existing and future code, ensuring repro...

SourceNational Centre of Competence in Research (NCCR) MARVEL·JournalNature Reviews Physics·TypeComputational simulation/modeling·DateNov 14, 2023

Predicting the optical read-out of a qubit from first principles

The study uses many-body perturbation theory to predict the optical properties of negatively charged boron vacancies in hBN, showing that phonons are largely responsible for luminescence. The results suggest that this defect can be used as a nanoscale thermometer with high temperature sensitivity.

SourceNational Centre of Competence in Research (NCCR) MARVEL·JournalPhysical Review Letters·TypeComputational simulation/modeling·DateApr 23, 2022

Researchers identify new paraelectric phase prototypes for use in computational engineering of functional materials

Researchers at NCCR MARVEL identified two new cubic prototypes that exhibit energetically and dynamically stable paraelectric behavior, providing a microscopic representation of the material's properties. The discovery has significant implications for the study of ferroelectricity, superconductivity, and other functional materials.

SourceNational Centre of Competence in Research (NCCR) MARVEL·JournalPhysical Review Research·TypeComputational simulation/modeling·DateMar 30, 2022

New paper describes on-surface synthesis and characterization of nitrogen-substituted undecacenes

Researchers successfully synthesized nitrogen-substituted undecacenes using on-surface chemistry, retaining their electronic properties with modified orbital energies. The study offers a new method for investigating complex electronic correlation effects in acenes and developing organic electronics and spintronics.

Equivariant representations for molecular Hamiltonians and N-center atomic-scale properties

Researchers develop a symmetrized N-center representation that provides a natural, fully equivariant framework for learning properties associated with multiple atoms. The approach gives excellent accuracy for predicting atomic properties despite using only linear or kernel regression.

SourceNational Centre of Competence in Research (NCCR) MARVEL·JournalChemical Physics·TypeComputational simulation/modeling·DateJan 10, 2022

Manifolds in commonly used atomic fingerprints lead to failure in machine-learning four-body interactions

Researchers found that two commonly used atomic fingerprints, ACSF and SOAP, are insensitive to certain movements, leading to the failure of machine learning in resolving four-body interactions. This limitation affects the accuracy of reproducing these interactions with limited success.

SourceNational Centre of Competence in Research (NCCR) MARVEL·TypeComputational simulation/modeling·DateJan 10, 2022

Machine learning solves the who’s who problem in NMR spectra of organic crystals

Researchers developed a machine learning method to assign NMR spectra of organic crystals probabilistically from their 2D chemical structures. The approach uses a database of chemical shifts for organic solids, reducing computational cost by up to 10,000 times compared to current methods.

SourceNational Centre of Competence in Research (NCCR) MARVEL·JournalScience Advances·TypeComputational simulation/modeling·DateNov 26, 2021

Newly identified R-2 2D material may show promise in development of spin-layer-locking spinFETs

Researchers at NCCR MARVEL identified lutetium oxide iodide (LuIO) as a high-performance material for spin-layer-locking spinFETs. They demonstrated the control of its properties with electric gates, providing practical guidelines for building and operating devices from this material.

SourceNational Centre of Competence in Research (NCCR) MARVEL·JournalNano Letters·TypeComputational simulation/modeling·DateSep 14, 2021

Common workflows for computing material properties with various quantum engines

Researchers have developed a common workflow interface for various quantum codes, enabling accurate predictions of system properties and promoting the wider use of density-functional theory. The interface allows users to optimize structures using any code without defining parameters, providing reusable results.

SourceNational Centre of Competence in Research (NCCR) MARVEL·Journalnpj Computational Materials·DateAug 23, 2021

Low-temperature crystallization of phase-pure α-formamidinium lead iodide enabled by study

A combined molecular dynamics and experimental study reveals a two-step process that enables the formation of phase-pure α-FAPbI3 at lower temperatures. The researchers used metadynamics to simulate the transformation from PbI2 to perovskite, which was confirmed by in situ x-ray and thin-film experiments.

New artificial neural network model bests MaxEnt in inverse problem example

A new artificial neural network model has been developed to solve inverse problems, demonstrating accuracy comparable to the maximum entropy (MaxEnt) approach. The model's versatility and robustness against noisy data have been showcased in various tests, including recovering electron single-particle spectral densities.

Scientists find self-healing catalyst for potential large-scale use in hydrogen production

Researchers have developed a self-healing catalyst, SION-X, that can efficiently release hydrogen from ammonia borane, a promising energy carrier. The catalyst is based on abundant mineral Jacquesdietrichite and can be easily regenerated, stored, and handled, making it suitable for large-scale applications.

SourceNational Centre of Competence in Research (NCCR) MARVEL·JournalJournal of Materials Chemistry A·DateApr 10, 2019

Researchers improve description of defective oxides with first principles calculation

The researchers developed site-dependent +U correction parameters using self-consistent first principles calculations, which improves the structure of stoichiometric SrMnO3 and provides a more accurate description of defective systems. This approach can lead to lower computational cost and more precise predictions of defect energetics.