Material design – towards reality via exascale computing

Solids, molecules and atoms vibrate even at T=0K around their equilibrium positions. These collective lattice motions (phonons) are responsible for many interesting phenomena like thermal expansion, superconductivity, etc. Due to the computational power of IT4Innovations HPC, these phenomena can be described at an atomistic scale, and thermal expansion and heat capacity of any materials can be determined directly from quantum-mechanical calculations without any empirical parameter. The research team models IR and Raman absorption spectra and mutual interaction of phonons to understand how heat is transported, and what it affects in matter. These calculations allow the design of novel nuclear fuels, searching for ultra-hard solids similar to diamond, as well as improving the figure of merit of thermoelectric materials. Another specific area of interest of our colleagues within this flagship is the study of phonon-electron and phonon-spin interaction. This allows superconductivity to be modeled. Moreover, for the very first time magnetic materials can be modeled under finite temperatures directly using quantum-mechanical calculations aiming to describe the behaviour of magnetocalorics (cooling with a magnetic field, and understanding ultrafast demagnetization by laser pulse for novel data-storage techniques.

The flagship research team also studies, models, and designs magnetic materials at different spatial and time scales for several technological applications such as magnetic recording, spintronics, electric motors, electric generators, magnetic actuators, biomedicine (magnetic hyperthermia), magnetic refrigeration, etc. They apply and combine modeling techniques to describe and understand magnetic materials accurately. Some of these approaches use Density Functional Theory (DFT) to calculate magnetic intrinsic properties at the microscopic scale and novel structures predicted by evolutionary algorithms, atomistic spin dynamics (to take into account finite-temperature effects), molecular dynamics (to study grain boundary phenomena and crystal phase stability), micromagnetics (to calculate magnetic domains, microstructure effects, and hysteresis loops at the macroscopic scale) and multiphysics finite-element models (to simulate magnet performance at operating conditions).

Within the research of clean energy technologies, new alloys for novel thermonuclear fusion reactors, such as ITER (International Thermonuclear Experimental Reactor) are designed using realistic, quantum-mechanical models for real simulation of First Wall alloys in reactor vessels, which are exposed to the heat of hydrogen plasma with a temperature of cca 200,000,000 °C, protected only by an external magnetic field (5T) keeping the plasma confined. Combining Quantum Mechanics, Statistical Physics, and High-Performance Computing, the used methods thus create a kind of a synergic modeling engine. Recently, surprising properties were obtained for 5-components equiatomic alloys stabilized by the entropy term at high temperatures with properties much different to those single element constituents. A new category of so called high (HEA), middle, and low entropy alloys has been instigated. Despite energies mostly being focused on the mechanical properties, magnetic behaviour cannot be omitted, as it is able to influence the phase stability and mechanical properties. The very first HEA is composed of Cr-Mn-FeCo-Ni being surprisingly paramagnetic at RT. Thus the objective is to determine the effect of the composition and substitutions on the magnetic character here.

Design and prediction of physical properties for realistic materials requires accurate calculation of large systems of atoms. Electronic structure calculations based, for example, on DFT became, in fact, the main computing method in materials science. However, standard DFT calculations deal with relatively small system sizes of a few hundred atoms, both for the purpose of computing time and memory requirements as computational intensity increases with the square and cube of the number of atoms. For more realistic systems with dislocations, interfaces, grain boundaries, random and diluted alloys, nano-particles, and biomolecules, however, it is necessa - ry to perform simulations of an order of several thousand to several million atoms. The CONQUEST code, utilizing the density matrix approach with a local basis, is a very efficient parallel method that provides a linear scaling solution. Such calculations are capable of an accurate description of the growth processes of nano-sized systems such as nano-particles, quantum dots, and nano-wires.

Another large-scale atomistic modeling method is the molecular dynamics approach based on interatomic potentials. For many applications, it allows physical properties behaviour to be determined under a given temperature (diffusion, melting tempera - ture, coexistence of different phases). In such cases, DFT calculations would again be prohibitively expensive, or even unfeasible due to running long simulations in the order of nanoseconds, even when using the CONQUEST code. Within the research, machine and deep learning methods are applied to design accurate interatomic potentials from DFT calculations.

The flagship researchers develop several codes written in the Fortran, Python, and Matlab/Octave programming languages. Below there is a selection of those that have already been published and made publicly available.

• PNADIS: This code is an automatic analyser of the Peierls-Nabarro stress and dislocation cores developed in MATLAB. This code calculates the dislocation core structure, Peierls stress, the pressure field around the dislocation core, and the solid solution strengthening of a crystal based on the Peierls-Nabarro model and other models derived therefrom.
• MAELASviewer: An online application for visualization and analysis of magnetostriction via a user-friendly interactive graphical interface.