The team studied three catalyst formulations built on a mixed TiO₂–Al₂O₃ support: a baseline iron-nickel catalyst (Fe–Ni/TiO₂–Al₂O₃), a cerium-promoted version (Fe–Ni–Ce/TiO₂–Al₂O₃), and a molybdenum- and cerium-containing variant (Fe–Mo–Ce/TiO₂–Al₂O₃). Each catalyst was tested in catalytic methane decomposition reactions at temperatures between 600 and 850 degrees Celsius, both under dry conditions and in the presence of steam, a variable that industrial reactors often cannot avoid. The comparison revealed a striking divergence in behavior. The Fe–Ni–Ce formulation achieved the highest methane conversion of the series, reaching 96 percent at 850 degrees Celsius, and sustained its performance over extended time on stream. The Fe–Mo–Ce catalyst, by contrast, started out highly active but deactivated more rapidly, a decline the researchers traced to the accumulation of heavily graphitized coke on its active surface.

The reason for this divergence lies in the distinct chemistry that cerium and molybdenum introduce to the iron-nickel active phase. Through an extensive characterization campaign combining X-ray diffraction, temperature-programmed reduction by hydrogen (H₂-TPR), temperature-programmed oxidation (TPO), Raman spectroscopy, and thermogravimetric analysis (TGA), the team showed that cerium dioxide acts as an oxygen reservoir within the catalyst. CeO₂ enhances the reducibility of the Fe-Ni phases and boosts the mobility of oxygen species through the material, a property long prized in ceria-based catalysis. Molybdenum behaves differently: it exists as highly dispersed MoOₓ species and substantially alters the redox properties of the system, shifting the way the catalyst exchanges oxygen during the reaction. These changes at the atomic scale cascade upward into macroscopic behavior, determining not just how fast methane is split, but what kind of carbon is left behind.

That last point is the heart of the study. When methane decomposes on a transition metal surface, the carbon it deposits is not a uniform nuisance; it comes in structurally distinct forms ranging from disordered, defective amorphous carbon to well-ordered sp² graphitic phases. Raman spectroscopy, which distinguishes these forms by the relative intensity of the disorder-induced D band and the graphitic G band, together with TGA, which measures how readily the deposited carbon burns off in oxygen, allowed the researchers to correlate carbon structure with catalytic performance. The correlations were unambiguous. Cerium promotes the formation of predominantly ordered sp² carbon structures and, crucially, facilitates the gasification of the defective carbon species that would otherwise encapsulate and poison the active metal particles. Molybdenum does the opposite, favoring the growth of more thermally stable, highly graphitized phases that resist removal.

This insight resolves a long-standing puzzle in the methane decomposition literature. Iron- and nickel-based catalysts have been known for decades to produce filamentous carbon, carbon nanofibers, and carbon nanotubes, and atomic-scale imaging studies have shown how carbon dissolves into the metal particle, diffuses through it, and precipitates as a filament, often lifting the nanoparticle off its support and keeping it alive. But the balance between the “good” carbon that grows as filaments and the “bad” encapsulating coke that kills the catalyst has been difficult to control. The Kazakh team’s work suggests that the redox chemistry of the promoter, and the presence or absence of steam, together govern which carbon morphology wins. Oxygen delivered from the ceria lattice can oxidize and remove defective carbon before it matures into pore-blocking coke, while MoOₓ species steer deposition toward graphitized deposits that, once established, are thermally stable and effectively permanent.

Steam proved to be a second, independent lever. In gasification chemistry, water is the classic reagent for burning carbon off metal surfaces, converting solid deposits to CO and CO₂. The researchers tested the catalysts both dry and in the presence of steam and found that steam influences both the type and stability of the carbon deposits, interacting with the redox properties of the promoters in ways their mechanistic model now captures. In a proposed coke formation mechanism, the authors argue that the structure of the carbon deposit is set by the interplay between the oxygen-delivery capacity of the promoter system and the availability of gas-phase oxidants such as steam. Where oxygen transfer is vigorous and defective carbon is rapidly gasified, the catalyst stays clean and functional; where carbon is allowed to graphitize undisturbed, the deposit consolidates and deactivation follows.

The practical implications run in two directions, and the authors are explicit that neither is inherently better. For operators whose goal is maximum, sustained hydrogen output, the Fe–Ni–Ce/TiO₂–Al₂O₃ catalyst is the clear choice: it pairs the highest conversion measured in the study, 96 percent at 850 degrees Celsius, with superior stability over time on stream, precisely because its ceria component keeps the carbon deposit from hardening into a fatal overlayer. For a laboratory or industrial partner seeking to co-produce graphitic carbon materials, whether for battery anodes, conductive additives, or other applications where ordered graphitic carbon commands a premium, the Fe–Mo–Ce formulation is attractive despite its faster deactivation, because it manufactures exactly the kind of thermally stable, graphitized carbon that commands value. The same chemistry, in other words, can be tuned toward either product simply by choosing the promoter and adjusting the atmosphere.

This tunability matters because hydrogen economics are brutal. Conventional steam methane reforming, the dominant hydrogen production route today, emits carbon dioxide as an unavoidable byproduct, roughly 9 to 12 kilograms of CO₂ per kilogram of hydrogen in unmitigated plants, and requires extensive downstream gas separation to purify the hydrogen. Catalytic methane decomposition sidesteps both problems: the reaction CH₄ → C + 2H₂ is mildly endothermic, yields a hydrogen stream that needs no CO₂ scrubbing, and locks the methane’s carbon into a solid. If that solid is a high-quality nanostructure rather than a worthless soot, the economics shift further, since carbon nanotubes and graphitic carbons can sell for orders of magnitude more per kilogram than the hydrogen itself. The Kazakh results feed directly into this “two products from one molecule” strategy by showing how to make the carbon saleable on demand.

The work also carries technical lessons for catalyst design beyond the specific Fe–Ni system. The TiO₂–Al₂O₃ mixed support provides both mechanical stability and electronic interaction with the active metal phases, and the characterization data show that the promoters do not act in isolation. XRD revealed the crystalline phases present in the fresh and spent catalysts, while H₂-TPR quantified how easily the metal oxides were reduced, a proxy for the catalyst’s readiness to activate methane. TPO profiles of the spent catalysts, showing at what temperature the deposited carbon oxidizes, provided a direct read on carbon reactivity that matched the Raman evidence: cerium-containing spent catalysts burned off carbon at lower temperatures, consistent with a higher fraction of defective, gasifiable species, while the Mo-containing samples required more aggressive conditions, consistent with graphitized deposits. The methodological point, that Raman and TGA data taken together can predict catalyst deactivation behavior, is itself a contribution that other groups can adopt.

The research was carried out with support from the Ministry of Science and Higher Education of the Republic of Kazakhstan under scientific project AP25793961, which is devoted to developing inexpensive, highly efficient composite materials for hydrogen and nanocarbon production from methane. It reflects a broader push among methane-rich nations to convert a fossil feedstock into a low-carbon energy carrier without waiting for expensive carbon capture infrastructure. The Almaty team’s contribution is a design principle rather than a single recipe: the redox character of the promoter, cerium for oxygen mobility, molybdenum for graphite-favoring conditions, and steam as a modulating agent, is the control knob that determines whether a methane decomposition catalyst lives long or produces graphite.

Whether the approach survives scale-up remains to be seen. Industrial methane pyrolysis demands catalysts that survive thousands of hours, tolerate sulfur impurities in natural gas, and allow continuous removal of accumulating carbon, and no laboratory result yet guarantees that. But the demonstration that a 96 percent methane conversion can be sustained on a ceria-promoted iron-nickel catalyst, and that the same platform can be redirected toward graphitic carbon synthesis by swapping in molybdenum, gives process engineers something they have largely lacked: a rational, mechanistically grounded way to choose their catalyst based on which product they want more. In a field where the hydrogen and the carbon have always pulled in opposite directions, that may be the most valuable result of all.

The study appears in Catalysis Letters as volume 156, article 271.

Methane pyrolysis involves decomposing methane into hydrogen gas and solid carbon, circumventing the generation of CO2 and thereby representing a cleaner alternative to conventional hydrogen production technologies. Central to this process are molten catalysts, which facilitate the high-temperature reaction by lowering activation energies and enhancing reaction rates. Despite their pivotal role, identifying molten catalysts that are both efficient and stable under reaction conditions poses a daunting scientific task. Molten catalysts operate within an immense, complex chemical landscape characterized by dynamic atomic structures and fluctuating active sites, making experimental discovery resource-intensive and time-consuming.

The newly developed platform, DigMethpy, emerges as a groundbreaking solution harnessing the power of artificial intelligence to navigate this intricate chemical design space. This digital catalysis platform integrates vast quantities of scientific literature, experimental data, computational chemistry simulations, machine learning algorithms, and insights from advanced large language models. By fusing these diverse information sources, DigMethpy constructs a dynamic and iterative discovery framework that continuously refines its catalyst predictions based on real-time validation data, thereby pushing beyond conventional trial-and-error methodologies.

Within DigMethpy’s extensive database lie over 40,000 meticulously curated data points derived from more than 500 peer-reviewed research articles and computational studies. These records encompass a wide array of molten metals, alloys, salts, and composite catalyst systems, facilitating a comprehensive understanding of catalyst behavior under methane pyrolysis conditions. By mining this data treasure trove, the platform identifies critical physicochemical descriptors that correlate with catalytic performance. Notably, it highlights atomic charge distributions, diffusion dynamics, and hydrogen adsorption properties as fundamental factors driving catalyst activity and selectivity.

The chemical intricacies of molten catalysts are particularly challenging due to their disordered atomic arrangements and fluxional active sites, which continuously evolve at reaction temperatures. DigMethpy addresses this complexity by employing sophisticated machine learning models capable of interpreting these dynamic features, thereby enabling accurate predictions of catalyst performance. The platform’s predictive prowess was exemplified in the development of multicomponent nickel-iron-based molten alloys exhibiting remarkable catalytic activity, showcasing the transformative potential of AI-supported materials design in the energy sector.

Beyond facilitating accelerated catalyst identification, DigMethpy signifies a paradigm shift in materials research by demonstrating the seamless integration of machine learning and natural language processing into scientific workflows. This synergy allows for automated literature synthesis and hypothesis generation, reducing human bias and leading to more objective and comprehensive exploration of materials design spaces. By leveraging these capabilities, researchers can rapidly iterate through candidate materials, testing and refining hypotheses digitally before committing to costly laboratory experiments.

The impact of DigMethpy is far-reaching, promising not only advancements in methane pyrolysis catalyst development but also broad applicability in the accelerated discovery of functional materials across various domains. Its closed-loop, autonomous discovery cycle embodies the future direction of scientific research, where AI agents work alongside human experts to unlock insights hidden within voluminous datasets. Such advances are critical for meeting global energy challenges, enhancing resource efficiency, and transitioning towards a low-carbon economy.

Hao Li, Distinguished Professor at Tohoku University’s Advanced Institute for Materials Research and founding editor of the journal AI Agents, emphasized the transformative potential of the platform: “By integrating experimental and computational knowledge with machine learning and natural language processing within a unified framework, DigMethpy accelerates the development of next-generation catalysts essential for sustainable hydrogen production and other green technologies.” This integrative approach sets a precedent for future AI-enhanced research endeavors, underscoring the importance of interdisciplinary collaboration in addressing complex scientific problems.

The team behind DigMethpy plans to continually expand the database, improve machine learning algorithms, and develop autonomous multi-agent systems capable of independently conducting catalyst discovery workflows. These improvements aim to further reduce discovery timelines and enhance predictive accuracy, ultimately facilitating industrial-scale applications. This iterative evolution embodies a shift towards increasingly intelligent and self-sufficient research ecosystems, empowering scientists with powerful computational tools to exploit the full potential of existing and emerging data.

Published on May 13, 2026, in the journal AI Agents, the study detailing DigMethpy’s development highlights the critical role of artificial intelligence in shaping the future of clean energy technologies. By harnessing advances in computational science and materials informatics, researchers are moving towards a new era where data-driven discovery dramatically enhances innovation speed and outcome reliability. As the global community intensifies efforts to combat climate change, such AI-powered platforms form the backbone of transformative sustainability strategies.

In addition to advancing the field of methane pyrolysis, DigMethpy’s success serves as a testament to the power of digital tools in catalysis and materials science. By bridging experimental and theoretical domains, the platform exemplifies a holistic approach to scientific exploration, empowering researchers to surmount current limitations in catalyst design. This synergy not only expedites the materials discovery process but also opens avenues for uncovering unanticipated phenomena, potentially leading to breakthroughs beyond hydrogen production.

As the energy sector increasingly prioritizes decarbonization and environmental stewardship, innovations like DigMethpy are essential for realizing practical, sustainable solutions. The detailed insights generated into molten catalyst behavior provide a roadmap for designing materials with tailored properties, optimized performance, and enhanced durability. This knowledge foundation paves the way for scalable hydrogen production technologies that can integrate seamlessly into future clean energy infrastructures, thereby supporting global climate goals and economic growth.

In summary, DigMethpy represents a pioneering AI-empowered platform that unleashes the potential of big data and machine intelligence in the discovery of molten catalysts for methane pyrolysis. By amalgamating computational models, literature mining, and experimental feedback into a cohesive digital ecosystem, the platform radically transforms catalyst research, delivering faster, more efficient, and data-driven pathways towards cleaner hydrogen production. This innovation not only addresses immediate scientific challenges but also heralds a new era of autonomous, intelligent materials development critical for sustainable technological advancement.

Subject of Research: Development of AI-driven platform for accelerated discovery of molten catalysts for methane pyrolysis

Article Title: DigMethpy: an AI-empowered digital catalysis platform for methane pyrolysis molten catalyst design

News Publication Date: 13-May-2026

Web References: http://dx.doi.org/10.20517/aiagent.2026.11

Image Credits: Zihao Cheng et al.

Keywords

Artificial intelligence, methane pyrolysis, molten catalysts, hydrogen production, catalyst design, machine learning, materials discovery, computational modeling, sustainable energy, digital catalysis, molten alloys, data-driven research