The global annual production of methanol is ~100 million tonnes, predominately from oil-derived sources using heterogeneous catalysts. Synthesis of methanol from CO₂, using sustainable hydrogen, is a promising strategy for clean methanol that has reduced carbon footprint and facilitates net zero emissions. Current Pd catalysts for CO₂ hydrogenation lack selectivity, particularly in avoiding methane production. This project took a theory-led design approach to create Pd-based alloy catalysts that improve the selectivity of CO₂ hydrogenation to methanol. Density Functional Theory (DFT) calculations were combined with Artificial Intelligence (AI) clustering approaches to identify surfaces with single-atom alloy compositions that favour specific reaction pathways; particularly, this was the strong adsorption of monodentate formate, with Pd-Nb, Pd-Mo, and Pd-W alloys potentially significantly improving CO₂ activation as they facilitate an Eley-Rideal mechanism. The outcomes of this work facilitate the future rational design of alloy catalysts for CO₂ hydrogenation.

Use of carbon dioxide to create a hydrogen storage vector, such as the liquid fuel methanol, is a promising route of achieving net zero emissions. Around 100 million tonnes of methanol are produced from oil-derived sources annually, with the majority synthesised using heterogeneous catalysts. [1] Pd catalysts are active for the CO₂ hydrogenation reaction, but these catalysts lack the desired selectivity, i.e., towards methanol rather than methane. Pd-based alloy catalysts could be potential enablers for improving the selectivity of the CO₂ hydrogenation reaction, as demonstrated by the near-zero methane selectivity observed in 1:1 PdZn alloys. [2] This project aim was therefore to accelerate catalysts discovery in this domain, by establishing a theory-led workflow for screening and designing catalytic alloy materials for CO₂ hydrogenation to methanol.

In the initial stages of the project, the ARCHER2 facility was vital for computational modelling using density functional theory (DFT). The results show that the CO₂ activation and hydrogenation to formate, a key intermediate facilitating methanol selectivity, is a crucial step in the reaction over the Pd-based low-index facets, as they are endergonic under both ambient and experimental conditions (ΔG > 0.5 eV). [3] The FCC Pd (100) surface was identified as a more suitable catalytic surface than (111) [4], as a submonolayer coverage of hydrogen can form under experimental [5] conditions, which leaves active sites available for the desired reactions. We further investigated the intermediates in the initial CO₂ hydrogenation to formate across various Cu, Pd, Zn, CuPd and PdZn surfaces, and this provided a theoretical basis for characterisation of the CO₂ hydrogenation mechanism over potential new catalysts.

The reactivity of the 1:1 CuPd alloy is similar to the Pd metal, which shows how the alloy can be used for efficient precious metal utilization. The dissociation of CO₂ to yield CO, a methane precursor, is less favourable on the 1:1 PdZn alloy facets than on pure Pd surfaces, which is thought to suppress methane productivity though not prevent the reaction completely. However, the CO adsorption is weak, with desorption more favourable than further hydrogenation to CH₄. [6] Considering further hydrogenation steps, the activation energy to initially hydrogenate CO₂ to formate is notably low on the body-centred tetragonal (BCT) PdZn (110) surface, and the transition state here resembles a monodentate formate intermediate, HCOO^m* (Figure 1). We therefore believe that the PdZn alloy surface offers enhanced reactivity as it enables an Eley-Rideal type mechanism, which does not require a pre-adsorbed CO₂ species but instead relies on CO₂ in the gas-phase reacting with readily available pre-adsorbed hydrogen. Furthermore, a linear correlation can be identified between the activation energy required for hydrogenation of CO₂ to formate, and the the adsorption energy (E_ads) of the product HCOO^m*, across the various investigated metal and alloy surfaces.

Figure 1. Monodentate formate intermediate (HCOO^m*) adsorbed at the 1:1 BCT PdZn (101) surface, with Pd and Zn presented as larger dark blue and light blue, and oxygen, carbon, and hydrogen presented in red, grey, and white, respectively.

As we believe the formation of this monodentate formate is key to the reaction selectivity sought, single atom alloy (SAA) surfaces based on Pd, Cu, and Zn, were tested to find strongly adsorbing HCOO^m* materials. A subgroup discovery (SGD) [7] artificial intelligence analysis was used, facilitated by partnership between Cardiff Catalysis Institute and Fritz Haber Institute, and benefitting from the extensive electronic structure calculations that had been completed on ARCHER2 previously. For HCOO^m* adsorbed on a range of SAA Pd (111), (100), (110) and (211) surfaces, doped with Co, Cu, Ga, Ir, Ni, Os, Pd, Pt, Rh, Ru or Zn, 13 candidate features relating to the single atom (SA) or the adsorption site of HCOO^m* were used to capture the SAA surfaces’ electronic and geometric properties. The combination of DFT and SGD analysis were able to show that the electron affinity (EA) of the single atom (SA) dopant, which is an intrinsic material property, can be used to predict SAs facilitating HCOO^m* adsorption. The analysis therefore allows us to predict that Pd-based alloys including Nb, Mo and W will significantly improve Pd’s capability of CO₂ activation, by facilitating greater the important Eley-Rideal type mechanism.

The ARCHER2 supercomputing platform was critical to the completion of the work, which based on our estimates would otherwise take ~300 years on a single CPU core. The workflow and results from this investigation can lead the way for theory-led rational design of alloy catalysts, in this case specifically targeting mechanistic aspects of CO₂ hydrogenation. The general workflow offers potential applicability to other reactions making use of alloy catalysts, accelerating catalyst discovery.

Figure 2. (a) E_ads distributions of the HCOO^m* data set (grey) and the desirable subgroup identified by SGD (blue). The orange and purple highlighting show the E_ads for the SAs tested after screening for candidate SAAs. (b) E_ads of HCOO^m* on SAAs with respect to EA_site, where the highlighted region defines the thresholds given by the SG rules.

Outcomes

New knowledge: The methods delivered a workflow for high-throughput screening of alloy CO₂ hydrogenation catalyst candidates.

New technology: The methods, through combination with experimental investigations, enable predictive design of new catalyst species (rather than reliance solely on materials simulation for verification).

New cross-disciplinary collaborations: Establishing cross-institutional international partnerships between experts in the Catalysis and Artificial Intelligence domains.

References

[1] Net Zero Aviation Fuels: resource requirements and environmental impacts policy briefing, The Royal Society, 2023. 80

[2] ACS Catal. 2022, 12, 9, 5371–5379

[3] Phys. Chem. Chem. Phys., 2022, 24, 9360-9373

[4] Surfaces of Pd-based alloys as catalysts for CO₂ activation and hydrogenation to methanol, 2023, PhD Thesis, Cardiff University

[5] Faraday Discuss., 2023, 242, 193-211

[6] J. Phys. Chem. Lett. 2020, 11, 18, 7672–7678

[7] ACS Catal. 2025, 15, 4, 2916–2926

Main investigators – Cardiff Catalysis Institute, Cardiff University, UK: Igor Kowalec, Lara Kabalan, Zhongwei Lu, Eimear McCarthy, David Willock, Richard Catlow, Andrew Logsdail.

Theory collaborators – NOMAD Laboratory, Fritz Haber Institute, Germany: Herzain Rivera, Luca M. Ghiringhelli, Lucas Foppa, Matthias Scheffler

Experimental collaborators – Cardiff Catalysis Institute, Cardiff University, UK: Naomi Lawes, Isla Gow, Sofia Mediavilla Madrigal, Nicholas Dummer, Michael Bowker, Graham Hutchings.

Funding – The authors are grateful for funding by the EPSRC Centre-to Centre Project (Grant reference: EP/S030468/1). IK acknowledges the Cardiff School of Chemistry for a PhD studentship award. ZL acknowledges funding by the China Scholarship Council. AJL acknowledges funding by the UKRI Future Leaders Fellowship program (MR/T018372/1). HIRA acknowledges Secretaría de Educación, Ciencia, Tecnología e Innovación de la Ciudad de México (SECTEI), for the support through their Postdoctoral Fellowship program (Agreement: SECTEI/107/2022). The authors acknowledge computational resources and support from: the Supercomputing Wales project, which is part-funded by the European Regional Development Fund (ERDF) via the Welsh Government; the UK National Supercomputing Services ARCHER2, accessed via membership of the Materials Chemistry Consortium, which is funded by Engineering and Physical Sciences Research Council (EP/L000202/1, EP/R029431/1, EP/T022213/1); and the Isambard UK National Tier-2 HPC Service operated by GW4 and the UK Met Office, and funded by EPSRC (EP/P020224/1).

Contact – Dr Igor Kowalec (KowalecI@cardiff.ac.uk), Prof. Sir Richard Catlow (CatlowR@cardiff.ac.uk), Dr Andrew Logsdail (LogsdailA@cardiff.ac.uk).