Artificial intelligence helps to identify correct atomic structures

Bayesian inference enhanced structure search facilitates accurate detection of molecular adsorbate configurations.
Illustration showing how camphor adsorbs to a copper surface
The stable structures of camphor on Cu(111) are identified in the minima of the 6D adsorption energy surface, which is modeled as a function of molecular orientation and translations.

Functional materials are commonly employed in emerging technologies, such as green energy solutions and new electronic devices. These materials are typically blends of different organic and inorganic components and have many advantageous properties for novel applications. To achieve their full potential, we need precise knowledge on their atomic structure. State-of-the-art experimental tools, such as atomic force microscopy (AFM), can be used to investigate organic molecular adsorbates on metallic surfaces. However, interpreting the actual structure from microscopy images is often difficult. Computational simulations can help to estimate the most probable structures, but with complex materials, accurate structure search is computationally intractable with conventional methods. Recently, CEST group has developed new tools for automated structure prediction using machine learning algorithms from computer science.

In this most recent work, we demonstrate the accuracy and efficiency of our own Bayesian Optimization Structure Search (BOSS) artificial intelligence method. With BOSS, we identify the adsorbate configurations of a camphor molecule on a Cu(111) surface. This material has been previously studied with AFM, but inferring the structures from those images was inconclusive. Here, we show that BOSS can successfully identify not only the most probable structure, but also eight stable adsorbate configurations that camphor can have on Cu(111). We used these model structures to better interpret the AFM experiments and concluded that the images likely feature camphor which is chemically bound to the Cu surface via an oxygen atom. Analyzing single molecular adsorbates in this way is only the first step towards studying more complex assemblies of several molecules on surfaces and subsequently the formation of monolayers. The acquired insight on interface structures can help to optimize the functional properties of these materials.

More details can be found in the following publication:

J. Järvi, P. Rinke, and M. Todorović, Detecting Stable Adsorbates of (1S)-camphor on Cu(111) with Bayesian Optimization, Beilstein J. Nanotechnol.  2020, 11, 1577-1589.

  • Published:
  • Updated:

Read more news

An artistic rendition of a photo colliding with other particles.
Studies Published:

Bridging gaps in reactor damage modelling and embracing diversity in academia

With a keen interest in electron-ion interactions, doctoral researcher Evgeniia Ponomareva describes her experience in the Nuclear Materials and Engineering (NuME) research group led by Assistant Professor Andrea Sand in the Department of Applied Physics.
Lara Ejtehadian, Patrick Rinke, and Ilari Lähteenmäki sitting with coffee mugs and smiling to the camera.
Awards and Recognition, Research & Art Published:

Aalto Open Science Award Winner 2023 - Aalto Materials Digitalization Platform (AMAD)

We interviewed the AMAD team, winners of the first Aalto Open Science Award.
Lohmann, a white female, stands in front of a lit seaweed sculpture. Laakso, a white male, seated in a metallic room with wires.
Appointments, University Published:

Sparks of curiosity: two professors share what they wonder about

Two newly tenured Aalto professors talk about what drives them
People at the campus
Cooperation, Research & Art Published:

Aalto University to host CESAER Task Force Openness of Science and Technology

Aalto University is proud to host the CESAER Task Force Openness of Science and Technology on 16–17 April 2024.