Density functional theory, or DFT, has been the workhorse of computational materials science for decades. Resting on the foundational theorems of Hohenberg and Kohn, which show that all ground-state properties of a many-electron system are uniquely determined by the electron density, DFT reduces the intractable many-body Schrödinger equation to a manageable single-electron problem through the Kohn–Sham formalism. In practice, the approach yields exactly the quantities that matter for energy applications: electronic band structures, densities of states, formation energies, adsorption energies, and the Gibbs free energies of hydrogen adsorption that underpin catalyst design. The catch is cost. A single DFT calculation for a complex oxide or perovskite can demand days of high-performance computing time, and exhaustively screening even ten thousand candidates can consume tens of thousands of CPU-hours. For the vast combinatorial space of possible energy materials, standalone DFT simply cannot keep pace.

Machine learning offers a way out. By training predictive models on databases of DFT-computed properties, researchers can estimate bandgaps, formation energies, and adsorption energies for new candidate structures orders of magnitude faster than direct simulation. Classical algorithms still hold their own in certain regimes: random forests and gradient boosting methods such as XGBoost and LightGBM perform robustly across benchmarks, while kernel-based approaches like Gaussian process regression are notably data-efficient and provide native uncertainty estimates, a valuable property when deciding which expensive calculations to run next. But the current accuracy frontier belongs to deep learning architectures that read crystal structures directly. Graph neural networks such as the Crystal Graph Convolutional Neural Network, MEGNet, SchNet, DimeNet++, CHGNet, and the equivariant MACE model represent atoms and bonds as nodes and edges in a graph, learning end to end from raw structure without hand-crafted descriptors and, in their equivariant variants, respecting the rotational and translational symmetries of physics itself.

The review is particularly clear-eyed about the practical trade-offs among these models. The three most accurate architectures, DimeNet++, CHGNet, and MACE, also carry the highest computational cost per prediction, and none of the leading graph models provides native, well-calibrated uncertainty quantification. That matters because uncertainty estimates are precisely what active learning pipelines use to decide which candidates deserve expensive DFT follow-up, so teams using graph networks typically bolt on ensembling or dropout-based proxies that are themselves imperfectly calibrated. Simpler crystal graph models scale most readily to campaigns screening hundreds of thousands to millions of candidates, while the classical low-data methods remain, somewhat counter-intuitively, the most portable across unfamiliar materials classes. Model choice, the author stresses, depends on the target property, training set size, and downstream needs, not on accuracy figures alone.

At the heart of the modern workflow sits the active learning loop, an iterative dance between statistics and quantum mechanics. A trained machine learning model screens a large candidate library, flags the predictions where its uncertainty is highest, and prioritises those structures for targeted DFT calculation. The new results are appended to the training set and the model is retrained, and the cycle repeats. Reported efficiency gains vary by domain, but case studies reviewed in the article describe reductions in the required number of experiments or calculations by roughly an order of magnitude. In one landmark campaign, Tran and Ulissi trained an active-learning surrogate across intermetallic surfaces spanning 31 elements, reaching adsorption energy mean absolute errors of just 0.17 to 0.18 electronvolts and guiding the identification of 258 candidate surfaces across 102 alloys for hydrogen evolution, along with 131 surfaces across 54 alloys for carbon dioxide reduction.

None of this would be possible without the open-data infrastructure that has grown up over the past decade. The Materials Project contains more than 154,000 inorganic compounds computed within a standardised PBE framework, the Open Quantum Materials Database offers thermodynamic data for over a million compositions, and AFLOW provides electronic and thermomechanical data for more than 3.5 million compounds, uniquely including phonon dispersions for lattice stability assessment. Workflow automation frameworks such as pymatgen, FireWorks, AiiDA, and Atomate orchestrate thousands of calculations across computing clusters while tracking provenance and recovering from errors. The review identifies this shared data commons as arguably the single most important enabler of data-driven discovery, transforming DFT from a bespoke per-study activity into reusable community infrastructure, while also cautioning that shared infrastructure concentrates shared data-quality problems.

The applications span the full clean-energy portfolio. For the hydrogen evolution reaction, machine learning models trained on DFT-computed hydrogen adsorption free energies exploit the volcano-plot framework rooted in the Sabatier principle, which locates the optimal catalyst where hydrogen binds neither too strongly nor too weakly. For photocatalytic water splitting, which requires a bandgap of roughly 1.8 to 3.1 electronvolts and band edges straddling the water redox potentials, machine learning-guided screening of 5,158 unexplored hybrid organic-inorganic perovskites identified six lead-free candidates with suitable bandgaps and room-temperature stability. In carbon dioxide reduction, active machine learning guided the experimental discovery of copper-aluminium electrocatalysts that convert CO2 to ethylene with the highest reported Faradaic efficiency for that reaction, validated by in situ X-ray absorption measurements. In batteries, a classifier trained on just 40 labelled materials screened 12,831 lithium-containing structures from the Materials Project and flagged 21 candidates for superionic conductivity, while machine learning interatomic potentials now simulate ionic transport at near-DFT accuracy.

The review also confronts the field’s growing pains with unusual candour. Machine learning models inherit every weakness of their training data: major databases overrepresent stable, well-known compounds while underrepresenting metastable phases, disordered alloys, and amorphous structures; inconsistent calculation parameters across aggregated sources introduce systematic offsets; and poorly converged structures inject label noise. The chronic underestimation of bandgaps by the standard PBE functional undermines photocatalyst predictions, with the more accurate hybrid and GW methods too expensive to generate large datasets, though transfer learning is emerging as a partial remedy. Graph networks remain black boxes, and uncertainty quantification is not a solved problem, with no single method uniformly well calibrated, especially in the out-of-distribution regime where genuine discovery happens. Most strikingly, the review highlights reproducibility concerns that have been raised about two of the field’s most famous demonstrations, the GNoME deep learning model that scaled discovery to 2.2 million predicted stable crystals and the A-Lab autonomous synthesis campaign, arguing that the community should release full structural and provenance data alongside every large-scale discovery claim as a norm rather than an exception.

Looking forward, the review maps out four directions that follow directly from these limitations. Autonomous materials-discovery laboratories will couple the active learning loop to robotic synthesis and characterisation platforms, extending the vision demonstrated by the A-Lab. Explainable AI, including physics-constrained neural networks and symbolic regression that extracts human-readable equations from trained models, will address the interpretability gap. Quantum computing, through variational quantum eigensolvers and quantum phase estimation, may eventually supply near-exact training labels for strongly correlated systems, although practical quantum advantage remains a multiyear to decade prospect. Multi-fidelity modelling will blend cheap, low-accuracy DFT data with sparse high-accuracy calculations to reconcile the accuracy-cost trade-off. Together with a fifth priority of independently re-executable benchmarking, these directions define the agenda for the coming decade.

The verdict of the review is ultimately optimistic but disciplined. Density functional theory provides the rigorous quantum mechanical foundation that gives machine learning models physical grounding, while machine learning supplies the statistical power to generalise those quantum insights across millions of unexplored candidates at negligible marginal cost. The two are most powerful not in isolation but as a closed, active learning loop, increasingly connected to robotic laboratories that can synthesise what the algorithms propose. Landmark studies, from the intermetallic electrocatalyst campaigns to the experimentally validated copper-aluminium CO2 catalysts, show that the framework already delivers real materials. Whether the clean energy revolution arrives on schedule, the author suggests, will depend less on the raw scale of these computational engines than on the discipline and verifiability with which they are extended to new chemistries.

Subject of Research: Integration of density functional theory and machine learning for accelerated discovery of energy materials

Article Title: Integration of density functional theory and machine learning for materials discovery in energy applications

Article References: Integration of density functional theory and machine learning for materials discovery in energy applications. (n.d.). https://doi.org/10.1007/s44371-026-00985-y

Image Credits: AI Generated

DOI: 10.1007/s44371-026-00985-y

Keywords: density functional theory, machine learning, materials discovery, energy materials, hydrogen evolution reaction, photocatalysis, CO2 reduction, battery materials, graph neural networks, active learning, high-throughput screening, Materials Project

The appeal of photothermal catalysis lies in its efficiency arithmetic. Most chemical processes demand either low-temperature heat below 150 degrees Celsius or medium-temperature heat between 150 and 400 degrees Celsius. Techno-economic studies identify heat pumps powered by renewable electricity as the cheapest option for the low-temperature range, but the medium-temperature band remains difficult to decarbonize. When sunlight is the primary renewable resource, the conventional route of converting photons to electricity and then electricity to heat suffers substantial losses; total light-to-heat efficiencies often fall below 20 percent for resistive heating and below 15 percent when electrolytic hydrogen is burned. Concentrated solar thermal systems, which convert sunlight directly into heat, can reach efficiencies of up to 60 percent. Photothermal catalysis pushes this logic one step further by eliminating the separate heat-transfer loop altogether: the catalyst itself absorbs the light and becomes the reactor’s heater.

Translating that concept from laboratory model systems to scalable technology has been the central obstacle. Many of the most celebrated photothermal catalysts in the literature are fabricated by colloidal lithography, self-assembly, or physical vapor deposition, producing highly regular nanostructures that are perfect for mechanistic studies but hold only vanishingly small catalyst volumes. Powder-supported alternatives scatter light so strongly that photons penetrate only tens of micrometers, leaving most of the catalyst bulk cold and idle. The Zurich team, led by researchers in the laboratory of Marcos Niederberger, reasoned that transparent aerogels could resolve this dilemma. Aerogels combine minimal light scattering with high surface area and open porosity, and they possess exceptionally low thermal conductivity, so heat generated deep inside the material stays there instead of leaking away. Previous work had incorporated metal nanoparticles into aerogels by co-gelation, but that route requires pre-synthesizing the nanoparticles and carefully controlling their dispersibility, which is difficult to scale. Standard wet impregnation, meanwhile, tends to destroy transparency or collapse the delicate pore network.

The new strategy hinges on a simple electrostatic trick performed in water. The team first synthesized zirconia nanocrystals of roughly three nanometers by a nonaqueous sol-gel route, then gelled them into uniform translucent spheres about a millimeter in diameter by extruding a partially gelled droplet through a syringe into heated silicone oil. Surface tension rounded each droplet into a glassy pearl before it solidified. Supercritical drying with carbon dioxide preserved the porosity, and calcination at 300 degrees Celsius removed residual organics. The resulting spheres retained about 92 percent porosity and a surface area of roughly 244 square meters per gram, far exceeding the 30 to 150 square meters per gram typical of conventional opaque zirconia powders. Because rewetting a dried gel collapses its pores, the metal had to be introduced before drying, and the researchers chose strong electrostatic adsorption using cheap, water-soluble palladium nitrate paired with ethylenediaminetetraacetic acid, or EDTA.

EDTA proved to be the linchpin of the method. At a pH of about 3.5, the zirconia surface carries a positive charge, while the Pd-EDTA complex is negatively charged, so the metal complex locks onto the gel surface electrostatically and survives the subsequent solvent exchange and supercritical drying. EDTA coordinates an unusually wide range of catalytically relevant metals and, crucially, prevents the hydroxide precipitation that plagues metals such as indium when ammonia-based complexation is attempted. Calcination cleanly burns the ligand away, and X-ray photoelectron spectroscopy confirmed that no nitrogen or significant carbon residues remained. Subsequent reduction in hydrogen generated metallic palladium nanoparticles with an average diameter of only about 1.5 nanometers, among the smallest values ever reported for supported palladium catalysts and well below the three to four nanometers typical of conventional impregnation or precipitation methods. The reduced spheres turned jet black, a visible sign that the palladium nanoparticles now absorbed light across the visible spectrum and could act as embedded nanoheaters.

Under concentrated white light from a high-power LED delivering 4.8 watts per square centimeter, roughly 48 suns, the catalyst converted a hydrogen and carbon dioxide feed into carbon monoxide through the reverse water-gas shift reaction. The researchers found that carbon monoxide production rose with palladium loading up to about 1.5 weight percent and then saturated, revealing a key design principle: once the nanoheaters absorb essentially all incident light, adding more metal yields no further heating and no further activity. The catalysts did deactivate, losing roughly 42 percent of their initial productivity within 16 hours on stream, a decline the team traced mainly to sintering of the palladium particles, which grew from about 1.5 to 3 nanometers, along with shrinkage and densification of the aerogel backbone. That densification is a deactivation pathway unique to photothermal operation, because a denser support conducts heat away faster and lowers the operating temperature by more than 10 degrees Celsius.

The most striking results came from adding indium. Co-depositing indium through the same EDTA route slowed deactivation dramatically, cutting it from about 42 percent to between 25 and 30 percent at indium loadings of one to two weight percent, and it pushed carbon monoxide selectivity above 99 percent, suppressing the parasitic formation of methane that would otherwise lock the carbon into a dead-end product. Electron microscopy suggested partial co-location of indium with palladium, and the authors attribute the improved selectivity to electronic and geometric effects: electron donation from indium weakens carbon monoxide adsorption on neighboring palladium atoms, while indium breaks up contiguous palladium ensembles that would otherwise bind carbon monoxide strongly enough to hydrogenate it all the way to methane. Over a 110-hour stability test, the optimized palladium-indium catalyst settled into a steady productivity of about 0.7 grams of carbon monoxide per gram of catalyst per hour, placing it at the upper end of reported photothermal reverse water-gas shift catalysts, which typically deliver between 0.1 and 0.6 grams per gram per hour.

Perhaps the most persuasive evidence for the aerogel concept came from a set of destruction experiments. The researchers took their intact spheres and ground them, gently in one case and thoroughly in ethanol in another, producing reference materials with identical chemistry but progressively destroyed architecture. The dry-ground aerogel reflected about 11 percent more light and ran roughly 10 degrees cooler, losing about 35 percent of its carbon monoxide productivity. The wet-ground sample, which resembled an ordinary powder catalyst, fared far worse: it ran about 44 degrees Celsius cooler than the intact spheres, a gap far too large to explain by reflectance alone, and produced only about 29 percent as much carbon monoxide. Cooling experiments confirmed that the collapsed materials dissipated heat faster, demonstrating that the aerogel’s twin gifts of transparency and ultralow thermal conductivity are what allow the entire catalyst bulk, not just its surface, to reach reaction temperature.

The broader implication is that photothermal catalysis need not remain a curiosity of precisely patterned model surfaces. By pairing a versatile aqueous impregnation chemistry with a support that lets light in and keeps heat trapped, the ETH Zurich team has outlined a scalable recipe for reactors in which sunlight both powers the process and supplies its carbon feedstock from recycled carbon dioxide. Challenges remain, particularly the gradual densification of the aerogel backbone and the residual sintering of the active metal, and the authors note that further compositional optimization is possible. But with a simple, inexpensive route to loading a wide range of metals into translucent, thermally insulating supports, and with productivity figures already competitive with the best photothermal systems reported, the path from glassy millimeter-sized pearls to industrial reactors in high-sunlight regions suddenly looks considerably shorter. For an industry searching for both clean heat and clean carbon, these self-heating spheres offer a rare two-for-one proposition.

Subject of Research: Development of translucent ZrO2 aerogel-supported Pd and PdIn nanoparticle catalysts for photothermal CO2 reduction via the reverse water-gas shift reaction

Article Title: ZrO2 Aerogel‐Supported Pd Nanoparticles for Photothermal CO2 Reduction

Article References: Kiwic, D., Räz, L., Tervoort, E., & Niederberger, M. (2026). ZrO 2 Aerogel‐Supported Pd Nanoparticles for Photothermal CO 2 Reduction. Advanced Science, 13(52), Article e76221. https://doi.org/10.1002/advs.76221

Image Credits: AI Generated

DOI: 10.1002/advs.76221

Keywords: photothermal catalysis, CO2 reduction, zirconia aerogel, palladium nanoparticles, reverse water-gas shift, strong electrostatic adsorption, solar-to-heat conversion, indium promotion, aerogel spheres, decarbonization, nanoparticle sintering, renewable process heat

The core problem the researchers tackled is one that plagues nearly all conventional microbial production platforms: during growth, a substantial fraction of the carbon a cell assimilates is ultimately released as biogenic carbon dioxide. In methanol-grown organisms, this loss occurs through central metabolic reactions, most notably the decarboxylation of pyruvate to acetyl-coenzyme A, a step shared by virtually every industrial chassis in use or under development, as well as through the oxidative branch of the ribulose monophosphate cycle that methylotrophs use to assimilate formaldehyde derived from methanol. Every molecule of carbon dioxide vented in this way represents feedstock that was purchased, delivered, and then effectively thrown away, capping biomass yields and undermining the economics of industrial methylotrophic bioprocesses. Reducing carbon loss at these junctions is therefore an established priority for developing low-oxidative, carbon-efficient biocatalysts.

The solution explored in the study exploits an enzyme known as phosphoketolase, or PKT, which cleaves the sugar phosphates fructose-6-phosphate and xylulose-5-phosphate into acetyl-phosphate plus either erythrose-4-phosphate or glyceraldehyde-3-phosphate. Acetyl-phosphate is valuable because it can be converted to acetyl-coenzyme A, the essential biosynthetic precursor, either in a single step by phosphotransacetylase or in two steps via acetate kinase and acetyl-CoA synthetase, entirely without passing through pyruvate decarboxylation. The side products, erythrose-4-phosphate and glyceraldehyde-3-phosphate, are recycled back into the ribulose monophosphate and pentose phosphate pathways. In effect, the enzyme provides a carbon-conserving detour around the cell’s biggest carbon dioxide leak. Previous work had shown that PKT-based pathways can dramatically enhance biomass yield and carbon conversion efficiency in several microbial hosts, including the methanotroph Methylotuvimicrobium buryatense 5GB1C, where overexpression of its native pktB gene substantially improved methane conversion efficiency.

Bacillus methanolicus emerged as an ideal test chassis for several reasons. It is a facultative ribulose monophosphate methylotroph that grows rapidly on methanol, doubling in roughly 1.4 hours under optimal conditions, faster than other native methylotrophs such as Methylobacterium extorquens AM1 at 3.6 hours, the yeast Pichia pastoris at 4.6 hours, and engineered Escherichia coli strains at 4.3 hours. It is also industrially interesting in its own right, capable of producing more than 50 grams per liter of glutamate under optimized conditions as well as lysine. Crucially, while B. methanolicus lacks any native phosphoketolase gene, it already encodes the downstream machinery needed to convert acetyl-phosphate to acetyl-CoA. That combination made it the perfect organism to test whether a methylotroph without native PKT genes could benefit from the carbon-conserving pathway, a question that had remained open.

To carry out the engineering, the team first built a suite of new genetic tools for the organism. They adapted an anhydrous tetracycline-inducible expression system, originally developed for Bacillus subtilis, by fusing the B. subtilis xylA promoter to the Tn10 tet operator and placing the tetR repressor under the control of a strong native B. methanolicus methanol dehydrogenase promoter. When validated with a green fluorescent protein reporter, the system proved highly tunable: fluorescence rose with increasing inducer concentration, and after induction with 0.8 micromolar anhydrous tetracycline, reporter expression climbed roughly 19-fold over five hours, from about 2,450 to about 46,000 normalized relative fluorescent units, before declining as free inducer was depleted. The researchers also characterized four native constitutive promoters of varying strengths, driving reporter expression from the tuf promoter at the highest levels, followed by pdxK, icd, and the much weaker dppE promoter, giving the community a graded palette of expression tools for this organism.

With the expression toolkit in hand, the team introduced a codon-optimized pktB gene from M. buryatense into B. methanolicus. Under inducible expression, the engineered strain produced 0.49 grams of dry cell weight per gram of methanol consumed, compared with 0.41 grams for the empty vector control, a 20 percent improvement in methanol-to-biomass yield. Constitutive expression from native promoters delivered comparable gains: the pdxK-pktB and icd-pktB strains each reached 0.52 grams of dry cell weight per gram of methanol consumed, and the tuf-pktB strain reached 0.50 grams, against a control baseline of 0.42 grams, improvements of roughly 18 to 24 percent. The pdxK-pktB strain, which showed the greatest yield enhancement without the most severe growth penalty, was selected for deeper analysis. Notably, all pktB-expressing strains displayed some growth defect relative to controls, a trade-off the researchers attribute to disruption of the carefully balanced flux through the ribulose monophosphate cycle.

Biochemical and molecular analyses confirmed the enzyme was doing its intended job. Whole-cell lysates from the pdxK-pktB strain produced 0.86 millimolar acetyl-phosphate in a hydroxamate activity assay, nearly double the 0.44 millimolar measured in the control strain, demonstrating functional PKT activity inside the thermophilic host. Reverse transcription PCR showed pktB transcripts at 13.13 times the level of the rpoB housekeeping gene during mid-log growth. Most importantly, when cultures were grown in sealed bottles and headspace gas analyzed by gas chromatography, the pktB-expressing strain released 0.44 moles of carbon dioxide per mole of methanol consumed versus 0.49 for the control, a 9 percent reduction in total biogenic carbon dioxide production, directly linking the yield gain to reduced carbon loss as hypothesized.

To eliminate the burden of maintaining plasmids, which themselves caused growth defects in control strains, the researchers deployed a recently developed temperature-sensitive chromosomal integration method, inserting the pdxK-pktB expression cassette into the chromosome by homologous recombination. The resulting integrated strain, BMGA3::pktB, grew more slowly and consumed methanol more slowly than the wild type, yet still delivered a 19 percent biomass yield enhancement, producing 0.82 grams of dry cell weight per gram of methanol consumed versus 0.69 grams for wild type. Carbon dioxide output fell 12 percent relative to wild type, from 0.49 to 0.44 moles per mole of methanol consumed. Intriguingly, pktB expression in the integrated strain was far lower than in the plasmid strain, just 0.16 times the housekeeping gene, suggesting that once enzyme expression exceeds a certain threshold, other factors such as substrate limitation, protein misfolding, or the enzyme’s thermolability may cap the achievable yield gain at around 24 percent.

The study also revealed unexpected shifts in byproduct metabolism. Whereas pktB overexpression in the enzyme’s native host M. buryatense had previously increased acetate excretion, the integrated B. methanolicus strain excreted no detectable acetate, while the wild type released roughly 3 millimolar. The authors propose two possible explanations: the slower growth rate of the engineered strain may simply reduce overflow metabolism, or carbon flux through acetyl-phosphate may be preferentially channeled toward acetyl-CoA rather than acetate. Distinguishing between these scenarios will require future metabolic flux analyses and intracellular metabolite measurements. Either way, the result hints that the engineered strain channels more carbon toward acetyl-CoA, potentially driving greater flux through the oxidative branch of a partially non-cyclic TCA cycle that fluxomic studies have shown operates at low levels during methanol growth.

Beyond the specific yield numbers, the work carries a broader lesson for metabolic engineering: the outcome of installing a carbon-conserving pathway depends not only on the enzyme’s catalytic properties but on the metabolic architecture of the host into which it is placed. This is, to the authors’ knowledge, the first report of introducing a phosphoketolase into a native methylotroph that lacks a PKT system, and the phenotypes observed differ meaningfully from those in the enzyme’s original host. The researchers suggest that adaptive laboratory evolution of the integrated strain could resolve the growth defect and optimize flux through the combined PKT-ribulose monophosphate network, while adding xylose utilization genes could boost regeneration of ribulose-5-phosphate and further support the pathway. As methanol, increasingly available from renewable electricity and captured carbon dioxide, gains traction as a feedstock, phosphoketolase-based carbon conservation stands out as a broadly applicable strategy for squeezing more product out of every carbon atom.

Subject of Research: Metabolic engineering of Bacillus methanolicus with heterologous phosphoketolase to enhance biomass yield from methanol and reduce CO2 loss

Article Title: Rewiring Carbon Metabolism in Bacillus methanolicus via Heterologous Phosphoketolase Expression Enhances Biomass Yield From Methanol and Reduces CO2 Loss

Article References: Rewiring Carbon Metabolism in Bacillus methanolicus via Heterologous Phosphoketolase Expression Enhances Biomass Yield From Methanol and Reduces CO2 Loss. (n.d.). https://doi.org/10.1111/1751-7915.70430

Image Credits: AI Generated

DOI: 10.1111/1751-7915.70430

Keywords: Bacillus methanolicus, phosphoketolase, methanol, methylotroph, RuMP cycle, carbon conservation, biomass yield, CO2 reduction, metabolic engineering, synthetic biology, biomanufacturing, acetyl-CoA

The core problem the researchers set out to solve is well known in the electrochemistry community. Copper remains the only metal catalyst that can convert CO2 into multi-carbon products at appreciable rates, because its binding energies sit in a narrow window that permits carbon–carbon coupling. Yet even the best copper-based systems overwhelmingly favor two-carbon products such as ethylene and ethanol, while the formation of C3+ hydrocarbons and oxygenates—propane, butane, propanol, butanol and other chemicals that command higher market value—remains stubbornly inefficient. Decades of catalyst design, from oxide-derived copper to facet-engineered films and tandem catalytic cascades, have delivered incremental gains, but the underlying reaction network on copper constrains which intermediates can form and, ultimately, which products can emerge.

The Yale–Antwerp team took a different approach: instead of redesigning the catalyst, they redesigned the feedstock. In their platform, CO2 or CO gas first passes through a non-thermal plasma, where energetic electrons collide with gas molecules and generate a rich cocktail of vibrationally excited molecules, radicals, dissociation fragments and unusual species such as carbon suboxide, C3O2. This activated gas stream is then delivered directly to a copper gas diffusion electrode, where the electrocatalytic reduction takes place. Crucially, by coupling the plasma to a gas-phase electrode rather than bubbling plasma products through an electrolyte, the design ensures that short-lived, highly reactive species survive long enough to reach the catalyst surface, where their stored chemical energy can be harvested electrochemically.

The results are striking. When CO2 and CO were co-fed through the plasma into the electrochemical cell, productivity of C3+ products increased by a factor of 3.3 and alcohol productivity by 1.5 times relative to electrocatalysis alone. Even more remarkable is the selectivity expansion: plasma activation unlocked the formation of chemicals that are essentially absent from purely electrocatalytic product distributions, including methanol, acetylene, ethane, propane, butane and butanol. In other words, the plasma does not simply accelerate the standard copper chemistry—it opens entirely new reaction channels on the same metal surface.

To understand why, the researchers combined plasma simulations, kinetic modeling and in situ spectroscopy. Plasma simulations of the discharge revealed that a significant fraction of CO2 molecules leaves the plasma vibrationally excited rather than fully dissociated. Vibrational excitation is a form of chemical currency: a vibrationally hot CO2 molecule effectively carries part of the activation energy needed for its own reduction, lowering the energetic barrier for subsequent steps at the electrode. Kinetic simulations comparing plasma-excited and ground-state species showed that these excited molecules, once adsorbed on copper, can follow different reaction trajectories than their thermal counterparts, enriching the population of key surface intermediates that feed C–C coupling and oxygenate formation.

Carbon suboxide emerged as another central player. This unusual C3O2 species, long studied in combustion and plasma physics, is generated in the plasma through reactions of CO with excited CO2. The team’s kinetic simulations and in situ analysis suggest that carbon suboxide arriving at the copper surface can be reduced and hydrogenated along pathways that bypass the conventional CO dimerization route, providing a direct entry point into three-carbon products. The in situ Raman spectroscopy experiments, performed by the Yale group on operating copper electrodes, tracked changes in surface adsorbates when plasma-activated gas was introduced, providing experimental evidence that the plasma species reshuffle the intermediate landscape on the catalyst rather than merely increasing local reactant concentration.

The gas diffusion electrode architecture deserves particular attention. Gas diffusion electrodes have become the workhorse of modern CO2 electrolysis because they bring the gaseous reactant into intimate contact with the catalyst at a three-phase boundary, enabling current densities that far exceed what submerged electrodes can achieve. The researchers deliberately designed the plasma outlet to feed the activated gas stream directly into the gas diffusion layer of the copper electrode, minimizing the residence time between plasma and catalyst. This tight coupling prevents the reactive plasma species from being consumed by recombination reactions or neutralized in the liquid electrolyte—a failure mode that has plagued earlier attempts to combine plasma with electrochemistry in solution-phase cells.

The team systematically varied the operating conditions to disentangle the contributions of different plasma species. Comparing pure CO2 feeds, pure CO feeds and mixed CO2–CO feeds through the plasma showed that the presence of both gases in the discharge enhanced C3+ and alcohol formation beyond what either gas alone could achieve, consistent with the formation of carbon suboxide requiring both CO2 and CO in the plasma zone. Varying the applied electrode potential mapped out how the electrochemical driving force interacts with the chemically activated feed, showing that the plasma effect persists across the potential window relevant to multi-carbon production. Plasma diagnostic experiments on the discharge itself complemented the modeling, anchoring the simulated species distributions in measured reality.

Beyond the mechanistic insights, the work carries significant implications for how renewable electricity might be converted into storable fuels and industrial feedstocks. Electrosynthesis of chemicals from CO2 promises a route to close the carbon cycle using intermittent solar and wind power, but the technology’s economic viability hinges on achieving high rates, high selectivity and high value products simultaneously. By demonstrating that plasma activation can raise production rates of the most valuable C3+ products by more than threefold while adding entirely new product classes, the platform addresses the selectivity bottleneck in a fundamentally new way—as a feedstock activation problem rather than a catalyst design problem. The two technologies are complementary by nature: plasma excitation is fast, catalyst-agnostic and tolerant of dilute feeds, while electrocatalysis offers precise control over electron transfer and product distribution at ambient temperature and pressure.

The researchers are candid that scaling the concept will require attention to energy efficiency, since both plasma generation and electrochemical reduction consume electricity, and the overall energy conversion chain must compete with incumbent fossil-based processes. Stability of the coupled reactor over extended operation, integration of the plasma unit with industrial gas handling, and optimization of discharge type and power coupling all represent engineering challenges ahead. A patent application covering the plasma–electrocatalysis flow cell has been filed by Yale, signaling the team’s intent to translate the concept. Still, the demonstration that vibrationally excited molecules and carbon suboxide can rewrite the reaction network on copper offers the field a genuinely new lever. If the synergy between plasma chemistry and electrocatalysis can be pushed further—through tailored discharge conditions, optimized gas residence times and catalyst surfaces designed to accept plasma species—the electrosynthesis of high-value chemicals and fuels from CO2 could move meaningfully closer to practical reality.

Subject of Research: Plasma-activated electrocatalytic conversion of CO2 and CO to C3+ hydrocarbons and oxygenates on copper

Article Title: Unlocking reaction pathways for CO2 and CO electrocatalytic reduction to C3+ hydrocarbons and oxygenates using plasma activation

Article References: Unlocking reaction pathways for CO2 and CO electrocatalytic reduction to C3+ hydrocarbons and oxygenates using plasma activation. (n.d.). https://doi.org/10.1038/s41929-026-01609-5

Image Credits: AI Generated

DOI: 10.1038/s41929-026-01609-5

Keywords: CO2 reduction, electrocatalysis, non-thermal plasma, copper catalyst, C3+ products, oxygenates, carbon suboxide, vibrational excitation, gas diffusion electrode, plasma-electrocatalysis, renewable fuels, Nature Catalysis