IAEA Oct. 3, 2018
A team of four scientists from the Max Planck Institute of Plasma Physics and from the Max Planck Computing and Data Facility in Garching, Germany have recently won the IAEA crowdsourcing challenge for visualization, analysis and simulation of materials to build fusion reactors.
Nuclear fusion, the reaction that powers the Sun, has the potential to eventually provide an unlimited supply of cheap and clean, carbon-free energy using isotopes of hydrogen obtained from water and lithium. However, harnessing commercially-viable fusion power involves serious technological challenges, including protecting the wall and other components of the reactor vessel from extremely high temperatures and high-energy particles.
Fourteen research teams from ten different countries submitted innovative analyses of simulations of the damage that can be caused to the reactor wall by the high-energy neutrons released by the fusion reaction. The simulations were judged on their scientific benefit, the novelty of the algorithm itself or its use within the domain of material science, and the utility and expected impact of the visualization.
“Some of the submissions were quite extraordinary, it is almost like organising a local football event and then have a World Cup winning team coming,” said Sergei Dudarev, Manager of the Materials Programme at the United Kingdom Atomic Energy Authority, and one of the initiators of the challenge.
The winning team, made up of Udo von Toussaint, Javier Dominguez, Markus Rampp and Michele Compostella, applied an existing technique from machine learning and data science for the first time to identify and classify structures of defect in the simulated damaged crystals.
“The solution opens up a new and productive way to automatically categorize defect structures and hence deduce, in a quantitative way, the commonalities and differences between materials,” explained Arjan Koning, Head of the Nuclear Data Section at the IAEA. “In the context of the study of materials for the vacuum vessel of a nuclear fusion reactor such as ITER, it provides an effective means of measuring, classifying and visualizing the damage done to a particular material by the high-energy neutrons released by the fusion reactor. The search for a suitable material from which to construct the reactor vessel’s first wall is a crucial step towards the construction of a viable fusion power plant.”
The approach has several advantages over existing methods:
- new or unexpected defect types can automatically be identified and classified;
- it is based on a combination of robust and well-understood algorithms from data science;
- it is able to distinguish between genuine defects and the small, temporary distortions caused by thermal movement of the atoms;
- it is fast enough to be applied during the evolution of the simulated damage in the crystal over time, to better understand how defects form, combine and (sometimes) eventually disappear as the atoms relax back to their initial positions on the crystal lattice.
Up to now, defect identification and classification were very labour-intensive and time-consuming tasks and, therefore, were typically carried out only at the end of molecular simulations. This new algorithm can be applied during the simulation of the crystal defect at each stage, which can provide new insights into when certain types of defects occur and vanish. This gives much more information about the system and allows identification of the types of defects that are likely to remain for a long time and those that are not – information which up to now was hardly accessible.
“We hope that our approach will tremendously accelerate the simulation analysis for molecular dynamics simulations,” said von Toussaint. “Computing power is increasing and manual capabilities are limited: anything that can be done by computer rather than by people, speeds up the scientific development.”
The winners will make their code available on a cost-free, open-source basis to any interested party, he added. It could be used by other institutions and experts – mainly material scientists – to analyse the results of their simulations, particularly those relating to radiation damage in solids.
The IAEA is planning to build on the success of this challenge by developing a distributed computing application that volunteers can download onto their computers to run simulations of damage in materials for fusion, Koning said. This has the potential to greatly increase the speed at which new candidate materials for a fusion reactor can be explored and will further enhance scientists’ understanding of the behaviour of these materials in such extreme conditions.
For the full, scientific description of the method, please see IAEA Challenge on Materials for Fusion: Winning Submission.