NCCR MARVEL scientists have conducted the most comprehensive validation effort to date on computer codes for materials simulation, providing colleagues with a baseline data set and guidance for evaluating and improving current and future codes.
Over the past several decades, physicists and materials scientists around the world have been busy developing computer codes that mimic the fundamental properties of materials. They can now choose from a full range of these tools and use them to publish tens of thousands of scientific articles. annually.
Understanding Density Functional Theory (DFT)
These codes are typically based on density functional theory (DFT), a modeling method that uses multiple approximations to reduce the incredible complexity of calculating the behavior of each individual electron according to the laws of quantum mechanics. The differences between results obtained using different codes depend on the numerical estimates made and the choice of numerical parameters behind these approximations, which are often adapted to study particular classes of materials, or to calculate key properties for specific applications – e.g. Conductivity of potential battery materials.
Code review challenges
Given the complexity of these codes, it is really difficult to ensure that they are all free from any possible coding error, or do not suffer from excessive numerical approximations. But it is crucial for the community to verify that the results of different codes are comparable, consistent with each other, and repeatable.
In a new article published today (November 14) in Physical nature reviewA large group of scientists has led the most comprehensive validation effort on solid-state DFT codes to date and has provided their colleagues with tools and a set of guidelines to evaluate and improve existing and future codes.
The work builds on an earlier study published in the journal Science in 2016 that compared 40 computational approaches each used to calculate the energies of a test set of 71 crystals, each corresponding to an element in the periodic table, and found that the traditional symbols were in very good agreement with each other. Some.
Expanding chemical diversity
“This work was reassuring, but it didn’t really explore enough chemical diversity,” says Giovanni Pizzi, head of the Program and Materials Data Group at the Paul Scherrer Institute PSI in Villigen, Switzerland, and corresponding author on the new article. “In this study, we considered 96 elements and simulated ten possible crystal structures for each. »
In particular, for each of the first 96 elements in the periodic table, they studied four different monoliths, which are crystals composed only of atoms of the element itself, and six different oxides, which also include oxygen atoms. The result is a dataset of 960 materials and their properties, which is calculated by two independent, state-of-the-art DFT codes called FLEUR and WIEN2k. Both are “all-electron” (AE) notations, meaning they explicitly take into account all electrons in the atoms under consideration.
Reference dataset for testing code
Anyone can now use this data set as a benchmark for testing the accuracy of other codes, particularly those based on pseudopotentials, where, unlike all-electronic (AE) codes, electrons not involved in chemical bonds are treated in a simplified way that makes the computation lighter.
“In fact, we have already started optimizing nine of these codes in our research, comparing their results with those in our data set, measuring deviations, and adjusting their numerical parameters (such as pseudopotentials) accordingly,” Pizzi explains.
Recommendations and future directions
The study also includes a series of recommendations for users of DFT codes to ensure reproducibility of computational studies, on how to use the reference dataset for future validation studies, and on how to expand it to other code families and beyond. Material properties.
“We hope that our dataset will set a standard in this field for years to come,” says Pizzi, one of nine MARVEL researchers who authored the study, along with Marnik Berks, Christian Emery, Sebastian Huber, Matthias Krack, and Nicola Marzari. , Alexander Yakutovich, Josung Yu, Austin Zadox.
Support IT frameworks
The study also provides an environment for future validation studies via AiiDA, the open access computing framework developed by the National Center for Research Competences (PRN) MARVEL, which supported the work and in which Pizzi is leading the project, and the European Research Centre. Max Center of Excellence. “AiiDA allows us to write the same instructions in the same way for 11 different pieces of code, for example a query to compute a particular structure,” Pizzi explains. It can then do the calculation for you and determine the correct numerical parameters for each.
In addition to expanding the standard data set with more structures, Pizzi says that in the future he hopes to take into account not only the accuracy of different codes, but also their cost in terms of time and energy calculations, thus helping scientists find the greatest cost. – Effective parameters of their calculations.
Funding: Swiss National Fund
“Music guru. Incurable web practitioner. Thinker. Lifelong zombie junkie. Tv buff. Typical organizer. Evil beer scholar.”