The development of stable Fe-based catalysts for CO2 hydrogenation to long-chain hydrocarbons faces a critical trade-off between selectivity and durability. Here, we demonstrate that ultra-high potassium loading (up to 30 wt %) on FeCu/gamma-Al2O3 catalysts modifying this limitation through electronic modification and mixed-phase stabilization. The optimized 25 wt% K catalyst achieves notable C5+ hydrocarbon selectivity of 52.6% while maintaining stable performance over 100 h. High-loading potassium promotion transforms the catalyst through: (1) electronic donation accelerating Fe5C2 formation, (2) enhanced surface basicity suppressing hydrogenation, and (3) stabilization of mixed Fe3O4/Fe5C2 phases resistant to deactivation. Although increased K induces carbon deposition, the deposited carbon is predominantly amorphous and carbide-type and is not associated with measurable activity loss within the test window. The gamma-Al2O3 support effectively disperses excess potassium, preserving catalyst porosity at ultra-high loadings. Long-term testing confirms sustained activity over 100 h with 15.6% C5+ yield and suppressed methane formation (12% selectivity), providing a design strategy for industrial CO2-to-fuel applications.
Social activities result from complex joint activity-travel decisions between group members. While observing the decision-making process of these activities is difficult via traditional travel surveys, the advent of new types of data, such as unstructured chat data, can help shed some light on these complex processes. However, interpreting these decision-making processes requires inferring both explicit and implicit factors. This typically involves the labor-intensive task of manually reading and annotating dialogues to capture context-dependent meanings shaped by social and cultural norms. Against this background, this study evaluates the potential of Large Language Models (LLMs) to automate and complement human annotation in interpreting decision-making processes from group chats, using data on joint eating-out activities in Japan as a case study. We designed a prompting framework inspired by the knowledge acquisition process in Knowledge Graph (KG) construction. This framework sequentially extracts key decision-making factors, including the group-level restaurant choice set and outcome, individual preferences of each alternative, and the specific attributes driving those preferences. This structured process guides the LLM to interpret group chat data step-by-step, converting unstructured dialogues into structured tabular data describing decision-making factors. To evaluate LLM-driven outputs, we conduct a quantitative analysis using a human-annotated ground truth dataset and a qualitative error analysis to examine model limitations. Results show that while the LLM reliably captures explicit decision-making factors, it struggles to identify nuanced implicit factors that human annotators readily identified. We further pinpoint specific contexts where the LLM performs poorly, establishing a boundary for when LLM-based extraction can be trusted versus when human oversight remains essential. These findings highlight both the potential and limitations of LLM-based analysis for incorporating non-traditional data sources on social activities into activity-based models.
Eliminating heavy metals from aqueous solutions continues to be of major environmental concern, primarily due to their non-biodegradable nature and hazardous effects. In this study, a Ca-modified low-temperature pine bark biochar (CaPB300) was developed as a sustainable and cost-effective adsorbent. The adsorption behavior of Cu (II) and Ni(II) in both single and binary systems was systematically investigated. Thermal treatment converted native carboxylic acid groups into deprotonated carboxylate (-COO-) groups, which acted as primary adsorption sites. The adsorption process was mainly governed by surface complexation, indicating a chemisorptiondominated mechanism. The Langmuir model demonstrated that in isolation, Cu(II) and Ni(II) demonstrated similar adsorption capacities at (56.85 and 55.88) mg/g, respectively. However, the introduction of a binary system significantly altered these figures to (65.20 and 11.45) mg/g for Cu(II) and Ni(II), respectively, confirming that when competing for available sites, Cu(II) outperforms Ni(II). In the binary system, Cu(II) was preferentially adsorbed over Ni(II) due to its stronger affinity toward carboxylate groups, leading to the formation of more stable surface complexes. Response surface methodology (RSM) was used to optimize the parameters for the adsorption process. The predicted removal efficiencies approached 100% for Cu(II) in both single and binary systems, while under optimized conditions, Ni(II) removal reached up to 99.39%. Overall, CaPB300 demonstrated efficient removal of Cu(II) and Ni(II), with pronounced selectivity toward Cu(II). These findings demonstrate the potential of CaPB300 as a cost-effective and sustainable adsorbent to treat multi-metal wastewater.
Ultrasonic testing offers a promising non-destructive approach for state-of-charge (SoC) estimation of lithium-ion batteries (LiBs). However, its practical deployment is hindered by domain shifts arising from variations in measurement conditions, particularly ambient temperature and sensor coupling conditions. Variations in these measurement conditions significantly alter macroscopic acoustic wave propagation—such as signal attenuation and baseline time-of-flight—producing distinct signal distributions at identical SoC levels. Existing methods mitigate this issue through multi-domain training or target-domain access, both of which are costly or infeasible in field deployment scenarios. In this work, we formulate single-source domain generalization as a practical problem setting for ultrasonic SoC estimation and propose SoC-Aware Contrastive Representation Learning (SCARL), a two-stage framework that learns domain-invariant representations from a single measurement condition. In the first stage, a frequency-domain preprocessing based on fast Fourier transform log-magnitude extraction suppresses condition-dependent phase variations, and a supervised contrastive learning objective with stochastic noise augmentation trains an encoder to cluster samples by SoC level while remaining invariant to measurement-condition-specific signal variations. In the second stage, the frozen encoder is coupled with a regressor to map the learned representations to continuous SoC values. Experimental results demonstrate that SCARL, trained exclusively on a single-source domain, achieves competitive SoC estimation accuracy on unseen domains, outperforming conventional approaches that assume access to multi-domain data. To the best of our knowledge, this is the first study to address single-source domain generalization in ultrasonic SoC estimation, enabling reliable cross-domain prediction without target-domain data collection or model recalibration.
Achieving Net-Zero emissions by 2050 and mitigating climate change demand efficient catalytic conversion of CO2 into valuable chemicals and fuels, a cornerstone for a sustainable circular carbon economy. This review comprehensively summarizes recent progress and critical catalyst properties in key thermal heterogeneous CO2 conversion processes, specifically dry reforming of methane, reverse water-gas shift, CO2-to-methanol synthesis, and the Sabatier reaction. For dry reforming of methane, overcoming coke deposition and metal sintering through optimized oxygen storage capacity, metal dispersion, and catalyst morphology is essential. In the reverse water-gas shift reaction, noble metal catalysts supported on reducible oxides exhibit superior intrinsic activity, but developing stable, cost-effective alternatives such as Mo-based transition metal carbides and dual-functional single-atom catalysts is imperative. CO2-to-methanol synthesis critically relies on precise engineering of metal–oxide interfaces and oxygen vacancies, with In2O3-based catalysts and strategic vacancy engineering demonstrating high potential. The Sabatier reaction, predominantly using Ni-based catalysts, requires improved catalytic stability and activity through advanced morphology control, incorporation of secondary metals, and alternative energy-assisted methods (e.g., plasma, mechanochemical).Overall, continued emphasis on catalyst–physicochemical property relationships, scalable synthesis, alternative energy integration, and techno-economic analyses will drive advances towards practical and efficient catalytic CO2 conversion within a circular carbon economy framework.
