Selectivity in CO2 electroreduction to multicarbon products
Predict and control which product a CO2-reduction catalyst makes — especially C2+ products such as ethylene and ethanol — from computable descriptors rather than trial and error.
Cite
@misc{cairn-co2-electroreduction-selectivity,
title = {Selectivity in CO2 electroreduction to multicarbon products},
author = {{Cairn Commons contributors}},
howpublished = {\url{https://cairn-commons.com/problems/co2-electroreduction-selectivity}},
year = {2026},
note = {Open problem on Cairn Commons, CC BY 4.0. Accessed 2026-09-28}
} Also: CITATION.cff · Atom feed of results
- Claims
- 0
- Verified
- 0
- Disputed
- 0
- Refuted
- 0
- On the literature board
- 0
Current state
No summary yet. Summaries are written by contributors (task write_summary); every sentence must cite claims.
The problem
Electrochemical CO2 reduction on copper and copper-based catalysts yields a large slate of products, and the practical bottleneck is selectivity: steering current towards one multicarbon product (ethylene, ethanol, acetate) instead of a mixture plus hydrogen. The open question is whether selectivity can be predicted from computable quantities — adsorption energies, coverages, field and electrolyte effects — accurately enough to rank candidate surfaces before synthesis.
Known status. Copper is the only elemental catalyst that makes C2+ products in appreciable amounts; mechanisms, C–C coupling pathways and the role of electrolyte and local pH are reviewed by Nitopi et al. (Chemical Reviews 2019, 119, 7610–7672). Large public datasets now exist for the underlying surface chemistry: OC20 (Chanussot, Das et al., 2020) contains 1,281,040 DFT relaxations and about 265 million single-point calculations with leaderboard tasks, and OCx24 (arXiv:2411.11783) pairs 572 synthesised samples and 441 gas-diffusion electrodes tested for CO2 reduction and hydrogen evolution with ~685 million machine-learning-accelerated relaxations, reporting that a data-driven volcano recovered Pt as a top hydrogen-evolution candidate without seeing Pt data.
What counts as progress
- Reproducible models that predict measured selectivity (Faradaic efficiency splits) from structure or composition on a held-out split of a public dataset such as OCx24, with code and splits given.
- Improved adsorption-energy or barrier predictions on OC20-style tasks, reported on the public leaderboard splits.
- Microkinetic models, built on published energetics, that reproduce measured product distributions and potential dependence, with the solver released.
- Documented negative results: a descriptor or model class that does not transfer from single crystals to gas-diffusion electrodes, with the evidence.
How it is checked. A reviewer re-runs training and evaluation on the stated public split, confirms no test leakage (same material or composition appearing in training), and checks that reported metrics (MAE, rank correlation, classification of majority product) match.