Source-linked AI summary
Single-Atom Alloy Catalysts Designed by First-Principles Calculations and Artificial Intelligence
Zhong-Kang Han, Debalaya Sarker, Runhai Ouyang, Yi Gao, Sergey V. Levchenko
TL;DR
The paper addresses the difficulty of rapidly and reliably predicting catalytic properties across many single-atom alloy candidates. It combines first-principles calculations with compressed-sensing descriptor discovery and reports models whose precision for segregation energy exceeds 95%, while screening broad candidate spaces.
Problem
Fast, reliable prediction of catalytic properties is needed because the large number of single-atom alloy candidates makes discovery difficult.
Method
The study combines RPBE-based first-principles calculations with SISSO descriptor selection and ab initio atomistic thermodynamics for catalyst screening.
Results
The models achieve precision higher than 95% for segregation energy and are evaluated using separate training and test sets with 10-fold cross-validation for descriptor dimensionality.
Takeaways & Limitations
Simple descriptors selected from large candidate spaces provide a basis for predicting hydrogen binding, H2 dissociation barriers, and segregation energy in single-atom alloy catalysts.
Abstract
from arXiv · showhide
Single-atom metal alloy catalysts (SAACs) have recently become a very active new frontier in catalysis research. The simultaneous optimization of both facile dissociation of reactants and a balanced strength of intermediates' binding make them highly efficient and selective for many industrially important reactions. However, discovery of new SAACs is hindered by the lack of fast yet reliable prediction of the catalytic properties of the sheer number of candidate materials. In this work, we address this problem by applying a compressed-sensing data-analytics approach parameterized with density-functional inputs. Our approach is faster and more accurate than the current state-of-the-art linear relationships. Besides consistently predicting high efficiency of the experimentally studied Pd/Cu, Pt/Cu, Pd/Ag, Pt/Au, Pd/Au, Pt/Ni, Au/Ru, and Ni/Zn SAACs (the first metal is the dispersed component), we identify more than two hundred yet unreported candidates. Some of these new candidates are predicted to exhibit even higher stability and efficiency than the reported ones. Our study demonstrates the importance of breaking linear relationships to avoid bias in catalysis design, as well as provides a recipe for selecting best candidate materials from hundreds of thousands of transition-metal SAACs for various applications.
Methods
The study combines RPBE-based first-principles calculations with SISSO descriptors and thermodynamic criteria to evaluate hydrogen binding, H2 dissociation barriers, and surface segregation in single-atom alloy catalysts.
- Computational setup: CI-NEB calculations identify transition-state structures for H2 dissociation barriers.The calculations use the RPBE functional implemented in FHI-aims.
- Energy definitions: BEH is calculated from the total H/support, alloy-support, and isolated-H energies.This definition uses equation (1).
- Energy definitions: SE measures the energy difference between placing a single impurity in the top layer and the nth surface layer with adsorbed H.The nth layer is selected when the energy difference from the preceding layer falls below 0.05 eV.
- Descriptor construction: SISSO selects simple, physically intuitive descriptors from more than ten billion candidates generated by combining primary features with mathematical operators.Feature spaces include up to three levels of operator complexity, with each higher space containing lower-level spaces.
- Candidate screening: Candidate selection uses ab initio atomistic thermodynamics to evaluate H adsorption and desorption as functions of temperature and hydrogen partial pressure.The free-energy treatment uses JANAF thermochemical data and fixes the hydrogen pressure at 1 atm in the described setup.
- Candidate screening: The acceptable BEH, Eb, and SE ranges are defined using free-energy, Arrhenius-type, and dopant-concentration criteria.The BEH criterion corresponds to |ΔG| < 0.3 eV, while acceptable barriers are below 0.3 eV at 298 K and SE bounds impose at least a 10% top-layer-to-subsurface dopant ratio.
Additional information
The paper provides publication and affiliation information alongside additional computational details for the first-principles calculations.
- Publication information: The paper lists Zhong-Kang Han, Debalaya Sarker, Runhai Ouyang, Yi Gao, and Sergey V. Levchenko as authors.Han, Sarker, Ouyang, and Levchenko are affiliated with institutions in Russia or China, while Gao is affiliated with the Chinese Academy of Sciences.
- Publication information: Four authors are identified as having contributed equally.The affiliations include Skolkovo Institute of Science and Technology, Shanghai University, and the Shanghai Advanced Research Institute.
- Computational details: The RPBE-based FHI-aims calculations achieve energy-difference convergence better than 10^-3 eV/atom and adsorption-energy basis-set superposition errors below 0.07 eV per molecule.Standard tight grids and basis functions are used, with spin polarization tested and included where appropriate.
S2. Additional SISSO computational details
The supplementary analysis constructs physically meaningful SISSO feature spaces from literature and DFT-derived primary features, then evaluates descriptor dimensionality and predictive reliability for BEH, Eb, and SE. It also documents the surface models, adsorption sites, screening maps, and validation data used for these quantities.
- Feature-space construction: SISSO feature spaces Φ1, Φ2, and Φ3 are generated by applying algebraic and functional operators to primary features.The operator set includes arithmetic, logarithmic, exponential, inverse, powers, roots, and absolute-value operations.
- Feature-space construction: Dimensional analysis retains only physically meaningful feature combinations, while primary features come from literature or DFT calculations.For example, features with incompatible units are not added or subtracted.
- SISSO optimization: The SISSO sparsifying ℓ0 constraint is applied after sure independence screening selects a smaller feature subspace, whose size depends on a user-defined SIS value.The supplementary text notes that optimizing this SIS value is not straightforward because searching the full feature space is computationally difficult.
- Model validation: Training and test data are separated by surface systems, with Pd(211) and Pt(111) systems reserved for testing and 10-fold cross-validation used to determine descriptor dimensionality.Descriptors are selected using the training data before evaluation on the held-out systems.
- Model validation: The selected SISSO models achieve precision above 95% for segregation-energy conditions, although SE deviations are larger than those for BEH and Eb.The precision criterion is evaluated using predicted and calculated segregation energies meeting the same SE < kTln(10) condition.
- Surface models and screening: The supplementary screening considers hydrogen adsorption on low-index and stepped transition-metal surfaces, with guest atoms substituted into host surfaces and multiple adsorption sites evaluated.The screened quantities include BEH, Eb, and SE, while the surface set includes fcc, hcp, and bcc facets.