Back

New Method of Analysis Developed by International Research Group Produces Accurate Information on the Structure of Carbon

The new methodology allows the experimental spectrum produced by X-ray spectroscopy to be separated into atomic-level data on the structure and surface chemistry of carbon. Figure: Anja Aarva / Aalto-yliopisto.

New Method of Analysis Developed by International Research Group Produces Accurate Information on the Structure of Carbon

Tommi Kutilainen

Researchers from Aalto, Cambridge, Oxford and Stanford Universities have taken a new step forward in describing the atomic nature of carbon-based materials. The long collaboration has its roots in the European Commission's HPC Europa3 transnational access programme.

Carbon-based materials are extremely versatile, but other elements are almost always present in them, which alters the materials' properties. Tailoring the materials to the needs of particular applications therefore requires atomic-level data on their surface structures and chemistry.

Detailed information on carbon surfaces can be obtained using X-ray spectroscopy. The spectrum it produces is challenging to interpret, however, because it summarizes information from several local chemical environments of the surface. This is where computational methods provide a new alternative, whose results are easier to interpret.

The researchers have developed a new systematic analysis method that uses machine learning to integrate the computational model with experimental results. The new methodology allows the experimental spectrum produced by X-ray spectroscopy to be separated into atomic-level data on the structure and surface chemistry of carbon.

– In the first phase, we identified atomic-level correspondences between spectra and each individual chemical environment. We then sought to aggregate the measured spectrum using computational spectrum reference data. With the help of machine learning, the computational spectra were grouped in a suitable way. Another method based on machine learning allowed us to produce a sufficient number of structures so as to get our hands on the non-crystalline structure of amorphous carbon, explains Anja Aarva, doctoral student at Aalto University and first author of the articles.

– In the past, interpretations of experimental results have varied based on who was examining them, but we were able to analyze our results using only computational reference data. Thanks to the new method, we now have a significantly better understanding of the surface chemistry of carbon, Aarva continues.

The researchers produced the computational spectra using CSC's Taito supercomputer and GPAW, a program based on density functional theory.

This study is a continuation of a 2017 HPC-Europa3 collaboration between Oxford professor Volker Deringer and his hosts at Aalto University, postdoctoral researcher Miguel Caro and professor Tomi Laurila. The study utilized machine learning methods developed by Deringer and professor Gabor Csányi from Cambridge University. Experimental measurements were carried out by Sami Sainio, an Aalto-based postdoctoral researcher at Stanford University.

HPC-Europa3 is a four-year European Commission program which funds transnational visits for European researchers and provides them access to supercomputers in nine participating European countries. CSC is responsible for coordinating HPC-E3 visits in Finland.
 
The study was published as a two-part article in the prestigious Chemistry of Materials journal.

Understanding X-ray Spectroscopy of Carbonaceous Materials by Combining Experiments, Density Functional Theory, and Machine Learning. Part I: Fingerprint Spectra

Understanding X-ray Spectroscopy of Carbonaceous Materials by Combining Experiments, Density Functional Theory, and Machine Learning. Part II: Quantitative Fitting of Spectra

This article is based on an Aalto University press release, "Tailor-made carbon can help scientists find hereditary diseases and the right doses of medication"

 



comments powered by Disqus