Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AUGUR, A flexible and efficient optimization algorithm for identification of optimal adsorption sites

Ioannis Kouroudis, Poonam, Neel Misciaci, Felix Mayr, Leon Müller, Zhaosu Gu, Alessio Gagliardi

TL;DR

A novel flexible optimization pipeline for determining the optimal adsorption sites, named AUGUR (Aware of Uncertainty Graph Unit Regression), which determines the optimal position of large and complicated clusters with far fewer iterations than current state-of-the-art approaches.

Abstract

In this paper, we propose a novel flexible optimization pipeline for determining the optimal adsorption sites, named AUGUR (Aware of Uncertainty Graph Unit Regression). Our model combines graph neural networks and Gaussian processes to create a flexible, efficient, symmetry-aware, translation, and rotation-invariant predictor with inbuilt uncertainty quantification. This predictor is then used as a surrogate for a data-efficient Bayesian Optimization scheme to determine the optimal adsorption positions. This pipeline determines the optimal position of large and complicated clusters with far fewer iterations than current state-of-the-art approaches. Further, it does not rely on hand-crafted features and can be seamlessly employed on any molecule without any alterations. Additionally, the pooling properties of graphs allow for the processing of molecules of different sizes by the same model. This allows the energy prediction of computationally demanding systems by a model trained on comparatively smaller and less expensive ones

AUGUR, A flexible and efficient optimization algorithm for identification of optimal adsorption sites

TL;DR

A novel flexible optimization pipeline for determining the optimal adsorption sites, named AUGUR (Aware of Uncertainty Graph Unit Regression), which determines the optimal position of large and complicated clusters with far fewer iterations than current state-of-the-art approaches.

Abstract

In this paper, we propose a novel flexible optimization pipeline for determining the optimal adsorption sites, named AUGUR (Aware of Uncertainty Graph Unit Regression). Our model combines graph neural networks and Gaussian processes to create a flexible, efficient, symmetry-aware, translation, and rotation-invariant predictor with inbuilt uncertainty quantification. This predictor is then used as a surrogate for a data-efficient Bayesian Optimization scheme to determine the optimal adsorption positions. This pipeline determines the optimal position of large and complicated clusters with far fewer iterations than current state-of-the-art approaches. Further, it does not rely on hand-crafted features and can be seamlessly employed on any molecule without any alterations. Additionally, the pooling properties of graphs allow for the processing of molecules of different sizes by the same model. This allows the energy prediction of computationally demanding systems by a model trained on comparatively smaller and less expensive ones
Paper Structure (11 sections, 8 equations, 8 figures, 3 tables)

This paper contains 11 sections, 8 equations, 8 figures, 3 tables.

Table of Contents

  1. Introduction
  2. AUGUR pipeline
  3. Density Functional Theory
  4. Bayesian Optimization
  5. Point generation
  6. Gaussian Processes
  7. Graph Neural Networks
  8. Results
  9. Chini Clusters
  10. Zinc Oxide Cluster
  11. Outlook

Figures (8)

  • Figure 1: AUGUR pipeline summary. The top line is the optimization pipeline, from left to right, training the GNNs and the GPs, using them as surrogates for BO, evaluating the BO suggestions with DFT, adding the new results into the dataset, and repeating them. The bottom line is the point generation for BO, from left to right, define the cluster, place the first adsorption atom at a physically consistent distance and outside the convex hull of the molecule, and repeat this process with all atoms of the adsorbant molecule.
  • Figure 2: Case Study 1: Chini clusters $\mathrm{[Pt_{3n}(CO)_{6n}]^{2-}}$ (n = 1–3); These nanoclusters comprise of three Pt-Pt bonds forming a triangle and laterally protected with CO ligands. As n increases, layers are progressively added, as seen in a), b), and c). The color scheme used is as follows: Pt (black); O (red); C (light grey).
  • Figure 3: Predicted energy surface for the Pt3 - Zn cluster in [eV] (left). The cluster figure depicts the most favorable adsorption position as predicted by AUGUR in the Pt$_3$ cluster (middle up). Uncertainty quantification of the prediction (right). The color scheme used is as follows: Pt (black); O (red); C (light grey), Zn (Blue)
  • Figure 4: Predicted energy surface for the Pt6 - Zn cluster in [eV] (left). The cluster figure depicts the most favorable adsorption position as predicted by AUGUR in the Pt$_6$ cluster (middle up). Uncertainty quantification of the prediction (right). The color scheme used is as follows: Pt (black); O (red); C (light grey); Zn (Blue)
  • Figure 5: Predicted energy surface for the Pt9 - Zn cluster in [eV] (left). The cluster figure depicts the most favorable adsorption position as predicted by AUGUR in the Pt$_9$ cluster (middle up). Uncertainty quantification of the prediction (right). The color scheme used is as follows: Pt (black); O (red); C (light grey); Zn (blue)
  • ...and 3 more figures