Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Design principles for III-nitride-nanocluster photocatalysts from region-resolved electronic structure

Shuaishuai Yuan, Gunther G. Andersson, Gregory F. Metha, Zetian Mi, Hong Guo

TL;DR

This work addresses how nanocluster cocatalysts modify III-nitride photocatalyst interfaces by introducing region-resolved electronic descriptors based on PLDOS to capture lateral heterogeneity between nanocluster-covered and uncovered surface regions. It combines a systematic first-principles dataset with a region-specific descriptor set and a physics-informed Interface Score to quantify interfacial efficiency, revealing that substrate-controlled electrostatics define the operating regime while nanocluster chemistry tunes local injection and coupling. Key contributions include the region-resolved PLDOS framework, the IS construction, and SHAP-based interpretability demonstrating how band bending, surface states, and interfacial charge redistribution govern hydrogen adsorption and charge transport. The approach provides transferable design principles and a generalizable workflow for engineering heterogeneous semiconductor–nanocluster interfaces in energy conversion applications.

Abstract

Understanding how nanocluster cocatalysts modify the electronic structure of III-nitride surfaces is central to the rational design of efficient photocatalytic interfaces. Here, we establish design principles for nanocluster cocatalysts on GaN-based semiconductors by systematically analyzing the spatially resolved electronic structure of GaN-, InGaN-, and ScGaN-based slabs decorated with six-atom elemental nanoclusters. Using a region-resolved projected local density of states (PLDOS) framework, we reveal that semiconductor-nanocluster interfaces operate as laterally heterogeneous electronic systems, in which nanocluster-covered regions govern charge injection and band bending, while uncovered nitride regions retain surface states that facilitate surface activation. We further show that cocatalyst effectiveness is controlled not only by hydrogen adsorption energy, but also by interfacial electrostatics, including band alignment, metal-induced gap-state suppression, and in-plane dipoles, with the semiconductor substrate defining the baseline electronic regime. Machine-learning regression models trained on physically motivated global and region-specific descriptors quantify the relative importance of these mechanisms and their correlation with hydrogen adsorption energetics. Together, this work provides transferable design principles for nanocluster cocatalysts on III-nitrides and a generalizable first-principles framework for studying spatially heterogeneous semiconductor-nanocluster interfaces.

Design principles for III-nitride-nanocluster photocatalysts from region-resolved electronic structure

TL;DR

This work addresses how nanocluster cocatalysts modify III-nitride photocatalyst interfaces by introducing region-resolved electronic descriptors based on PLDOS to capture lateral heterogeneity between nanocluster-covered and uncovered surface regions. It combines a systematic first-principles dataset with a region-specific descriptor set and a physics-informed Interface Score to quantify interfacial efficiency, revealing that substrate-controlled electrostatics define the operating regime while nanocluster chemistry tunes local injection and coupling. Key contributions include the region-resolved PLDOS framework, the IS construction, and SHAP-based interpretability demonstrating how band bending, surface states, and interfacial charge redistribution govern hydrogen adsorption and charge transport. The approach provides transferable design principles and a generalizable workflow for engineering heterogeneous semiconductor–nanocluster interfaces in energy conversion applications.

Abstract

Understanding how nanocluster cocatalysts modify the electronic structure of III-nitride surfaces is central to the rational design of efficient photocatalytic interfaces. Here, we establish design principles for nanocluster cocatalysts on GaN-based semiconductors by systematically analyzing the spatially resolved electronic structure of GaN-, InGaN-, and ScGaN-based slabs decorated with six-atom elemental nanoclusters. Using a region-resolved projected local density of states (PLDOS) framework, we reveal that semiconductor-nanocluster interfaces operate as laterally heterogeneous electronic systems, in which nanocluster-covered regions govern charge injection and band bending, while uncovered nitride regions retain surface states that facilitate surface activation. We further show that cocatalyst effectiveness is controlled not only by hydrogen adsorption energy, but also by interfacial electrostatics, including band alignment, metal-induced gap-state suppression, and in-plane dipoles, with the semiconductor substrate defining the baseline electronic regime. Machine-learning regression models trained on physically motivated global and region-specific descriptors quantify the relative importance of these mechanisms and their correlation with hydrogen adsorption energetics. Together, this work provides transferable design principles for nanocluster cocatalysts on III-nitrides and a generalizable first-principles framework for studying spatially heterogeneous semiconductor-nanocluster interfaces.
Paper Structure (31 sections, 25 equations, 12 figures, 7 tables)

This paper contains 31 sections, 25 equations, 12 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Results
  3. Overview of the dataset and model systems
  4. From PDOS to spatially resolved descriptors
  5. Region-Resolved Descriptors and Interface Score Construction
  6. Data-Driven Feature Attribution and Physical Interpretability
  7. Discussion
  8. Conclusion
  9. Methods
  10. DFT calculations
  11. Machine learning methods
  12. Global Descriptors
  13. Calculation of $R_{\mathrm{VBM}}$, $R_{\mathrm{Gap}}$, and $R_{\mathrm{CBM}}$
  14. Calculation of $R^{\mathrm{s}}_{\mathrm{VBM}}$, $R^{\mathrm{p}}_{\mathrm{VBM}}$, and $R^{\mathrm{d}}_{\mathrm{VBM}}$
  15. Calculation of s/p/d-band properties
  16. ...and 16 more sections

Figures (12)

  • Figure 1: Schematic of the constructed ab initio database comprising GaN, InGaN, and ScGaN slabs decorated with six-atom elemental nanoclusters spanning the periodic table. The framework can accommodate additional variations such as facets, alloying, oxidation or defect states, and water adsorption. Upon relaxation, the nanoclusters exhibit three structural motifs: attached (marked in green), deformed (marked in yellow), and dissociated (marked in dark gray). Elements shown in light gray do not form stable six-atom nanoclusters on GaN-based surfaces.
  • Figure 2: From PDOS to region-specific PLDOS for a six-atom Rh nanocluster on a GaN (110) slab with a single hydrogen atom adsorbed on the nanocluster. (a,b) Side and top views of the relaxed structure, defining Region A (beneath the nanocluster) and Region B (uncovered surface). (c) Bulk GaN band structure and PDOS reference. (d) PDOS of the Rh-decorated slab, illustrating overall density-of-states overlap but limited spatial resolution. (e) Layer-resolved projected local density of states (PLDOS) revealing band alignment and charge redistribution across the interface. (f,i–iv) Local PDOS of representative Ga and N atoms: (f,i) bulk-layer Ga, (f,ii) bulk-layer N, (f,iii) surface-layer Ga, and (f,iv) surface-layer N. (g,h) Region-resolved PLDOS for Regions A and B, normalized per orbital channel to emphasize spatial variations across layers. (i,j) PLDOS with element-wise normalization highlighting dominant orbital contributions in each region.
  • Figure 3: (a) Global descriptors capture composition-averaged electronic and chemical properties, including orbital band centers and widths, filling factors, work function, ionization energy, electron affinity, electronegativity, and Bader charge (see Table \ref{['tab:global']}). (b) Region-resolved descriptors distinguish the nanocluster-covered region (A) from the uncovered surface (B), quantifying band bending, nanocluster-induced gap states, surface-state peak energies and widths, in-plane dipole moments, lateral band-edge offsets, and interfacial charge redistribution (see Table \ref{['tab:iface']}). The schematic summarizes the spatial partitioning and functional roles of these descriptors.
  • Figure 4: (a,b) Parity plots comparing machine-learning predictions with DFT-calculated hydrogen adsorption energies $E_\mathrm{H}$ using (a) global descriptors and (b) region-resolved interface descriptors across multiple regression algorithms. Red and blue markers denote training and test sets, respectively; dashed lines indicate perfect agreement. (c,d) SHAP analysis of the best-performing models for (c) the global-descriptor set and (d) the region-resolved descriptor set, highlighting the most influential features and their directional contributions to $E_\mathrm{H}$. Feature color denotes relative feature magnitude. (e,f) Principal-component analysis (PCA) projections of (e) the global descriptors and (f) region-resolved descriptors, colored by substrate (GaN, InGaN, ScGaN), illustrating the enhanced substrate-specific separation captured by spatially resolved interface descriptors.
  • Figure 5: Element-resolved PDOS of Rh and H on the GaN slab, highlighting the surface state positions ($E_{\mathrm{VBM}}$ and $E_{\mathrm{CBM}}$) used to define integration windows. The computed ratios for Rh are: $R_{\mathrm{VBM}} = 0.5112$, $R_{\mathrm{Gap}} = 0.4017$, and $R_{\mathrm{CBM}} = 0.0871$. For H they are: $R_{\mathrm{VBM}} = 0.6736$, $R_{\mathrm{Gap}} = 0.1954$, and $R_{\mathrm{CBM}} = 0.1309$.
  • ...and 7 more figures