High-throughput computational-experimental screening protocol for the discovery of bimetallic catalysts

The study addresses the challenge of efficiently discovering catalysts by tightly integrating computational screening with experimental validation. Traditional first-principles approaches (e.g., full reaction pathway and barrier calculations) are accurate but too time-consuming for high-throughput screening. The authors hypothesize that materials with electronic density of states (DOS) patterns similar to a known active catalyst (Pd) will exhibit comparable catalytic performance. Building on concepts like the d-band center and band shape descriptors, they propose using the full DOS pattern (including sp and d contributions) as a simple, physics-grounded descriptor to rapidly bridge computational predictions and experimental catalyst testing, targeting replacement or reduction of Pd in direct H2O2 synthesis.

Prior work established correlations between d-band center and adsorption energies, enabling activity predictions via volcano relationships. Extensions involving higher moments of d-band shapes and sp-band properties improved reactivity capture for alloys. Studies showed that alloys with electronic structures similar to active noble metals can mimic their catalytic properties (e.g., Ir50Au50 vs Pt for H2 dissociation; Rh50Ag50 vs Pd for hydrogen storage). Despite these insights, full DOS patterns had not been used directly as a descriptor in computational-experimental screening workflows. The authors leverage this gap by adopting the complete DOS pattern similarity to Pd as the key descriptor for bimetallic catalyst discovery.

Computational high-throughput screening: - Space of candidates: 30 transition metals (periods IV–VI) combined into 435 binary systems at 1:1 composition; for each, 10 ordered crystal structure prototypes were considered (B1, B2, B3, B4, B11, B19, B27, B33, L10, L11), totaling 4350 structures. - DFT setup: Spin-polarized calculations with VASP using PAW pseudopotentials and the RPBE functional (with D3 added for mechanism studies). Plane-wave cutoff 400 eV; k-point meshes 8×8×8 (bulk) and 4×4×1 (slabs; 3×3×1 for mechanism). Bulk cells 1×1×1; slabs 2×2 with 4 layers and 15 Å vacuum. Relaxations to <1e-4 eV/cell and forces <0.01–0.02 eV/Å. - Thermodynamic screening: Formation energies ΔE = E[AB] − E[A] − E[B] computed for all 4350 phases; phases with ΔE ≤ 0.1 eV considered feasible (to allow nanosize-stabilized nonequilibrium alloys). This reduced to 249 alloy systems. - DOS similarity descriptor: For each thermodynamically admissible alloy, the close-packed surface facet was modeled (L11→(111), B2→(110), others→(001)). The surface DOS was computed as the sum of partial DOSs of surface atoms (including s, p, d). DOS similarity to Pd(111) was quantified as ΔDOS2−1 = {∫[DOS2(E) − DOS1(E)]^2 g(E;σ) dE}^{1/2}, with Gaussian weighting g(E;σ) centered at EF and σ=7 eV to emphasize energies around EF encompassing typical d-band centers (−3.5 to 0 eV). - Rationale for including sp-states: Illustrated via O2 adsorption on Ni50Pt50(111), where sp DOS changed significantly upon adsorption while d DOS did not, indicating key sp–p interactions. Candidate selection and synthesis feasibility filtering: - From 249 alloys, 17 candidates with low ΔDOS2−1 (<2.0) were shortlisted: CrRh (1.97, B2), FeCo (1.63, B2), CoNi (1.71, L10), CoHf (1.89, L10), CoTa (1.96, L10), CoPt (1.78, L10), NiCu (1.11, L10), NiNb (1.96, L10), NiMo (1.87, L10), NiRh (0.98, L10), Ni-Pd (0.84, L10), NiIr (1.28, L10), Ni-Pt (1.16, L10), CuPd (1.51, B2), RhPt (1.67, L10), Pd-Pt (1.16, L10), PdAu (1.43, L10). CoHf and CoTa were excluded due to element scarcity; NiCu and CrRh were excluded due to unfavorable reduction potential mismatches for wet-chemical alloying. - Nanoparticle synthesis: For the remaining 13 systems, nanoparticles (NPs) were synthesized via n-butyllithium reduction in dioctyl ether/oleylamine with controlled heating. Intended 50:50 loading produced ICP-verified compositions: Au51Pd49, Ni61Pt39, Pd50Cu50, Pd52Ni48, Pt58Co42, Pt52Pd48, Rh56Ni44, Rh56Pt44; CoFe, NiIr, NiMo, NiNb, and NiCo did not alloy under these conditions. - Characterization: XRD confirmed alloy phases and matched DFT-simulated patterns; Rietveld analyses indicated cubic-based structures, with minor non-alloyed Pt coexisting in Ni–Pt (alloyed Ni72Pt28 present; catalytic tests used Ni61Pt39 by ICP). STEM-EDX mapping showed homogeneous mixing; NP mean sizes: 3.4±0.4 nm (Pd50Cu50), 8.8±2.0 (Ni61Pt39), 5.0±1.5 (Rh56Ni44), 11.6±2.0 (Au51Pd49), 5.2±1.5 (Pd52Ni48), 8.1±2.2 (Pt52Pd48), 5.8±1.0 (Pt58Co42), 1.7±0.2 (Rh56Pt44). - Catalytic testing: Direct H2O2 synthesis at 20 °C, 1 atm for 1 h in ethanol/water (1:4) with 0.9 mM NaBr and 0.02 M H3PO4; 1 mg catalyst; gas flow 22 mL/min at H2:O2=1:10. Measured H2O2 productivity (mass- and cost-normalized), H2 conversion, H2O2 selectivity, and TOF. - Mechanism calculations: DFT energy profiles for Langmuir–Hinshelwood pathway and heterolytic electron–proton transfer (computational hydrogen electrode, implicit solvation) on Ni50Pt50(111) and Pd(111); CI-NEB for transition states; pCOHP and charge density difference analyses to probe bonding and back-bonding interactions. Composition optimization: - Experimentally varied Ni/Pt ratios in NixPt100−x NPs (structure stable across compositions) and identified Ni61Pt39 as optimal for productivity, H2 conversion, and selectivity.

- High-throughput outcome: 4350 structures screened → 249 thermodynamically feasible alloys (ΔE ≤ 0.1 eV) → 17 low-DOS-dissimilarity candidates (ΔDOS2−1 < 2.0) → 13 targeted for synthesis → 8 successfully alloyed as NPs → 4 exhibited catalytic properties comparable to Pd in direct H2O2 synthesis. - Catalytic performance (mass-normalized productivity, mmol H2O2 gmetal−1 h−1): Au51Pd49: 505.2; Ni61Pt39: 880.1; Pt52Pd48: 1225.6; Pd52Ni48: 218.1; Pd (benchmark): 552.4. - Cost-normalized productivity (mmol H2O2 $metal−1 h−1): Ni61Pt39: 69.3 (~9.5× Pd’s 7.3); Au51Pd49: 7.1; Pd52Ni48: 5.5; Pt52Pd48: 23.0. - H2 conversion (%) and H2O2 selectivity (%): Pd: 14.0% / 63.0%; Ni61Pt39: 22.9% / 61.2%; Au51Pd49: 11.6% / 70.1%; Pt52Pd48: 25.0% / 78.7%; Pd52Ni48: 2.8% / >99.9% (driven by low conversion). - TOF (h−1): Ni61Pt39: 98.5; Pd: 58.8; Pt52Pd48: 186.9; Au51Pd49: 77.1. - Descriptor validation: Alloys with lower ADOS2−1 generally showed better activity; ADOS2−1 differentiated performance better than d-band center differences in selected comparisons (e.g., Pt52Pd48 vs Pd50Cu50). - New catalyst discovery: Ni–Pt identified as a previously unreported catalyst system for direct H2O2 synthesis; Ni61Pt39 delivered superior cost-normalized activity while maintaining selectivity comparable to Pd. - Mechanistic insights: On Ni50Pt50(111), overall H2O2 formation via Langmuir–Hinshelwood is more exothermic (−0.42 eV) than on Pd(111) (−0.07 eV). Rate-determining step differs: Pd(111) first hydrogenation (I→II) barrier 0.78 eV; Ni50Pt50(111) H2O2* desorption (III→V) barrier 0.69 eV. Electron–proton transfer pathway also thermodynamically favorable on both, with slightly more favorable second protonation on Ni–Pt (−1.58 eV vs −1.40 eV). pCOHP and charge density analyses indicate stronger stabilization/back-bonding of H2O2* on Ni–Pt compared to Pd. - Composition optimization: Among NixPt100−x tested, Ni61Pt39 provided the best combination of productivity, conversion, and selectivity.

The study demonstrates that full DOS pattern similarity to a known active catalyst (Pd) is an effective, computationally light descriptor to predict Pd-like catalytic behavior without costly reaction pathway calculations. This approach efficiently narrows a vast alloy space to experimentally tractable candidates, leading to validated catalysts for direct H2O2 synthesis. The discovery of Ni–Pt as a new, cost-effective catalyst with performance and selectivity comparable to Pd substantiates the hypothesis that similar electronic structures yield similar catalytic properties. Moreover, the DOS descriptor outperforms simpler single-parameter metrics (e.g., d-band center) in ranking catalytic activity for this reaction. Mechanistic DFT analyses rationalize the observed performance, showing more favorable thermodynamics and different rate-limiting steps on Ni–Pt, and elucidating stronger H2O2* stabilization via back-bonding. These results underscore the descriptor’s relevance and the protocol’s capacity to accelerate catalyst discovery while addressing cost constraints by reducing reliance on platinum-group metals.

A computational-experimental protocol using full DOS pattern similarity as the screening descriptor efficiently identified bimetallic catalysts capable of replacing or reducing Pd usage for direct H2O2 synthesis. From 4350 screened structures, eight candidate alloys were synthesized, and four matched or exceeded Pd’s catalytic performance. Notably, Ni61Pt39 delivered superior cost-normalized productivity and comparable selectivity, marking the first report of Ni–Pt as an active catalyst for this reaction. The methodology provides a generalizable framework: by updating the reference DOS (e.g., to the best-performing Ni–Pt), the search can be iterated to discover further non-PGM or reduced-PGM catalysts and potentially extended to other reactions.

- Composition scope: Initial computational screening was restricted to 1:1 stoichiometries; optimal compositions were determined experimentally, indicating that broader compositional searches could further improve performance. - Structural models: Ordered phases were used as reference structures; real nanoparticles may include disordered/random solid solutions, potentially causing deviations between predicted and observed behavior. - Surface facets: Only close-packed surfaces were considered; catalytic behavior can be facet-dependent. - Synthesis constraints: Several predicted candidates (e.g., CoFe, NiIr, NiMo, NiNb, NiCo) could not be alloyed under the chosen wet-chemical conditions, limiting experimental validation. - Descriptor specifics: The DOS similarity metric depends on choices like Gaussian width and energy window; while effective here, transferability to other reactions may require recalibration. - DFT approximations: Use of RPBE and neglect of explicit solvent/finite-temperature dynamics in most screening steps may introduce errors in energetics; mechanistic steps were refined with dispersion and implicit solvation but still approximate.