Yuichi Yamaguchi

The infrared (IR) spectra of BiVO4 surfaces are sensitive to local atomic structure and can potentially provide insight into surface heterogeneity observed in recent spatially resolved nano-Fourier-transform infrared (nano-FTIR) spectroscopy measurements. However, no clear relationship has been established yet between IR spectral features and the surface atomic structures in BiVO4. Here, we used first-principles calculations to investigate the IR signatures of four low-index BiVO4 surface orientations with stoichiometric and Bi-rich terminations. We found that the most distinctive spectral feature is the symmetric stretching mode of VO43− units, which is present only on stoichiometric surfaces and exhibits facet-dependent frequency shifts and intensity variations. Analysis of the vibrational modes showed that these changes originate from differences in the magnitude and orientation of surface-induced V–O bond distortions relative to the surface normal direction. Comparison with experimental nano-FTIR spectra shows a consistent trend in the change of the symmetric stretching peak intensity across different surface facets, while discrepancies in the relative intensity ratio between asymmetric and symmetric stretching modes suggest additional contributions from experimental factors, including probe depth and tip orientation. Oxygen evolution reaction calculations further confirm the strong influence of surface orientation and termination on catalytic activity. These results establish a direct link between surface atomic structure, IR signatures, and catalytic behavior in BiVO4, providing a framework for interpreting nanoscale IR spectroscopy measurements of heterogeneous photocatalyst surfaces.

The optimal solid-solution ratio of the stannite-type CuxAg2–xZnGeySn1–yS4 photocatalyst for sacrificial H2 evolution under visible light (λ ≥ 420 nm) was predicted using a machine learning approach. The prediction mapping indicated that the region around (x, y) = (0.6, 0.3) was the optimal solid-solution ratio. Based on the mapping, we experimentally investigated the activities of the CuxAg2–xZnGeySn1–yS4 photocatalyst, and the ratio around (x = 0.8, y = 0.6) was found to be optimal. Notably, the formation of the solid solution guided by the machine learning prediction enabled both a response to visible light with long wavelength and improvement in the activity of the CuxAg2–xZnGeySn1–yS4 photocatalyst compared with those of the previously developed Ag2ZnMS4(M = Ge and Sn) photocatalyst with a stannite structure. Furthermore, we found that the training data with compositions similar to that of CuxAg2–xZnGeySn1–yS4 played an important role in generating the prediction mapping. This finding suggests that compositionally neighboring data outside the target plane can be effectively leveraged to enable reliable data-driven materials discovery even when data within a specific composition plane are severely limited. This study is expected to promote the development of multielement functional materials beyond photocatalysts.

In particulate photocatalysis, typical experiments involve billions to trillions of particles that possess highly heterogeneous properties, making it difficult to uncover mechanisms and structure-property relationships at the single-particle level. Additionally, existing single-particle measurement techniques become limited by the need to anchor particles to substrates and lack the ability to control interparticle spacing and interactions. Herein, we describe a multi-modal technique called optical particle trapping with electrochemical sensing (OPTECS), whereby the distributions of product generation rates from fully wetted, individual photocatalyst particles are measured using microelectrode sensors while simultaneously recording their luminescent characteristics. Using doped bismuth vanadate and strontium titanate as model particulate photocatalysts towards overall water splitting, OPTECS is used to reveal (i) hidden correlations between single particle luminescence intensity and photoactivity, and (ii) the presence of optically stimulated electrochemiluminescence (OSECL), which is leveraged to quantify rates of undesired back reactions, a key loss mechanism in particulate photocatalysis.

Optimal solid-solution ratio of stannite-type Cu _x Ag _{2- x} ZnGe _y Sn _{1- y} S ₄ photocatalysts for sacrificial H ₂ evolution under visible light ( λ ≥ 420 nm) was predicted using a machine-learning approach. The Cu _x Ag _{2- x} ZnGe _y Sn _{1- y} S ₄ plane (hereafter referred to as the "in-plane"), defined as a two-dimensional composition space spanned by x and y , consists of 231 points when x and y are 0.0–2.0 and 0.0–1.0 with 0.1 increments, respectively. H ₂ activities at only six data points of ( x, y ) = (0.0, 1.0), (0.5, 1.0), (1.0, 1.0), (1.5, 1.0), (2.0, 1.0), and (2.0, 0.0) within the in-plane that located at the edge of the plane had been previously measured. Prediction mapping using our original machine-learning method designed to address data imbalance and experimental-condition dependence in photocatalyst datasets indicated that the region around ( x, y ) = (0.6, 0.3) was predicted to exhibit the optimal solid-solution ratio with maximal photocatalytic activity. On the basis of the prediction mapping, we experimentally investigated the activities of Cu _x Ag _{2- x} ZnGe _y Sn _{1- y} S ₄ photocatalysts with various solid-solution ratios. The ratio around ( x =0.8, y =0.6) was found to be optimal, and the highest photocatalytic activity was experimentally confirmed in this central solid-solution region. The optimized photocatalyst responded to 660 nm and the apparent quantum yield (AQY) was 8.2% at 420 nm. It was notable that response to visible light with long wavelength and improvement in the activity of the Cu _x Ag _{2- x} ZnGe _y Sn _{1- y} S ₄ photocatalyst compared to that of previously developed Ag ₂ ZnGeS ₄ with a stannite structure were achieved by formation of solid-solution based on the machine-learning prediction. Furthermore, we examined the contribution of out-of-plane data to the prediction mapping, where out-of-plane data were defined as data points that were included in the training dataset used to construct the machine-learning model but were not represented by the Cu _x Ag _{2- x} ZnGe _y Sn _{1- y} S ₄ . The sequential removal of subsets of out-of-plane data from the training data and the sequential regeneration of the predicted activity maps demonstrated that out-of-plane data compositionally similar to Cu _x Ag _{2- x} ZnGe _y Sn _{1- y} S ₄ played an important role in generating the prediction mapping. Composition-space analyses based on dimensionality-reduction, clustering, and distance analyses further revealed that the out-of-plane data located close to the Cu _x Ag _{2- x} ZnGe _y Sn _{1- y} S ₄ compositions contributed significantly to capturing activity trends within the in-plane. These insights suggest that compositionally neighboring data outside the target plane can be effectively leveraged to enable reliable data-driven materials discovery even when data within a specific composition plane are severely limited. This study is expected to promote the development of multi-element functional materials beyond photocatalysts.

Ag ₂ CdGeS ₄ and Cu ₂ CdMS ₄ (M = Ge and Sn) predicted by machine learning were experimentally developed for sacrificial hydrogen evolution under visible light.

Hybrid photocathodes combining metal sulfide powders with poly-3,4-ethylenedioxythiophene (PEDOT) as an effective hole transporting polymer were prepared through a facile drop-casting method and electrochemically oxidative polymerization of 3,4-ethylenedioxythiophene (EDOT). The PEDOT modification remarkably enhanced a photocurrent density of a powder-based (CuGa)_0.5ZnS₂ photocathode (bandgap (BG): 2.3 eV) for hydrogen evolution under visible light irradiation at 0 V vs RHE, 60-times higher compared with an unmodified photocathode. The improvement was advantageously effective when using small metal sulfide particles prepared by a flux method compared to large particles synthesized by a solid-state reaction due to improved necking and interfacial contact facilitated by PEDOT in the fine physical structure. The optimized PEDOT/(CuGa)_0.5ZnS₂-composited photocathode (PEDOT-(CuGa)_0.5ZnS₂) achieved an incident photon-to-current conversion efficiency (IPCE) of 30% at 420 nm and 0 V vs RHE for the H₂ evolution. PEDOT modification was also effective for Cu₂ZnSnS₄ (BG: 1.4 eV) and Cu₃VS₄ (BG: 1.5 eV) black metal sulfide photocathodes. Furthermore, solar water splitting proceeded in a two-electrode photoelectrochemical cell consisting of a PEDOT-(CuGa)_0.5ZnS₂ photocathode and a BiVO₄-based photoanode, even without applying an external bias. The photoelectrochemical cell gave 0.08% for nonbias and 0.21% for 0.4 V of an external bias of a solar-to-hydrogen energy conversion efficiency (STH), even though both photoelectrodes were fabricated without any vacuum process and using just two ingredients. Our findings will offer a viable strategy for developing photocathodes that unite simplicity and high performance by using powder materials for the scalable application of photoelectrochemical green H₂ production.

Large language models (LLMs) have the potential to serve as collaborative assistants in scientific research. However, adapting them to specialized domains is difficult because it requires the integration of domain-specific knowledge. We propose Materials Dual-Source Knowledge Retrieval-Augmented Generation (MDSK-RAG), a retrieval-augmented generation (RAG) framework that enables domain specialization of LLMs for materials development under fully offline (no-Internet) operation to ensure data confidentiality. The framework unifies two complementary knowledge sources, experimental CSV data (practical knowledge) and scientific PDF literature (theoretical insights), by converting tabular records into template-based text, retrieving relevant passages from each source, summarizing them with a local LLM, and merging the summaries with the user query prior to generation. As a case study, we applied the framework to metal-sulfide photocatalysts using 740 in-house experimental records and 20 scientific PDFs. We evaluated the framework on a benchmark consisting of 14 expert-defined questions and used two-sided Wilcoxon signed-rank tests for paired comparisons. Models with fewer than 10 billion parameters were executed on a laptop, whereas larger models were run on a dedicated local server; the cloud-based LLM (GPT-4o) was evaluated via the cloud service. For practical deployment, gemma-2-9b-it (W = 14.0, two-sided p-value: p = 1.34 × 10^-2) and improved the median expert 5-point rating from 2 to 3, an absolute increase of 1 point (corresponding to a relative percentage gain of 50.00%; Wilcoxon signed-rank test statistic: W = 3.5, two-sided p-value: p = 7.00 × 10^-3). For reasoning-type questions, incomplete context retrieved by MDSK-RAG sometimes disrupted the model's reasoning process and led to incorrect conclusions, indicating remaining room for improvement. Comparable, statistically significant improvements were observed for the other local models (Qwen2.5-7B-Instruct and a larger gemma-2-27b-it) between conditions with and without the framework in the evaluation by cosine similarity to expert reference answers. In comparison to a cloud-based LLM, the gemma-2-9b-it with the framework outperformed GPT-4o. In this case study, the framework effectively incorporated practical experimental knowledge and theoretical literature into local LLM responses, improving accuracy for domain-specific queries. The framework presented here offers a practical and extensible adaptation of local LLMs to domain-specific scientific research.

Operando X-ray absorption spectroscopy enabled real-time monitoring of valence-state changes in a water-soluble ionic cobalt complex electron mediator during aqueous Z-scheme CO ₂ reduction in a semiconductor/metal-complex hybrid photocatalysis.

Large language models (LLMs) have garnered increasing attention owing to their potential as collaborative assistants in scientific studies. However, adapting an LLM to specialized domains remains challenging because of the difficulty in incorporating domain-specific knowledge. In the present study, we propose a framework named materials dual-source knowledge retrieval-augmented generation (MDSK-RAG). The present framework is a variant of RAG frameworks that enables domain adaptation of a local LLM without fine-tuning and is applied to metal-sulfide photocatalyst studies. Experimental results stored in a CSV file and scientific literature written as PDF files were used as sources of domain-specific knowledge by the two retrievers of the present MDSK-RAG. The tabular data were converted into natural scientific language according to a predefined template to improve their interpretability. The texts retrieved from both sources were summarized before response generation using the local LLM. The summary is performed by the LLM based on a predefined prompt that extracts information corresponding to five aspects of metal-sulfide photocatalysts: composition, crystal structure type, synthesis method, reaction conditions, and hydrogen evolution activity. The evaluation of the generated responses to 12 expert-defined questions revealed that the present MDSK-RAG improved the scientific quality of the responses in terms of both factual correctness and expert-specific reasoning. The proposed framework can operate entirely in a local environment and offers advantages in terms of data confidentiality. In addition, a user-friendly application was developed to support its practical use in materials development workflows. The proposed framework can be generalized and adapted to a broad range of materials research domains.

A photoelectrochemical cell as artificial photosynthesis is an attractive system for solar CO₂ reduction in H₂O to produce a mixture gas of H₂ and CO, so called “green syngas.” Development of photocathodes with the response to long wavelength visible light and high selectivity for CO formation is necessary in terms of efficient solar energy conversion. Recently, we reported that a powder-based Cu₃VS₄ black photocathode was active for solar H₂ production utilizing a whole range of visible light.¹ This photocathode was a Cu(Ⅰ)-containing metal sulfide, which was a suitable material for solar CO₂ reduction.² Therefore, solar CO₂ reduction using a Cu₃VS₄ photocathode is expected to be achieved by constructing a photoelectrochemical cell. In the present study, we developed a new photoelectrochemical cell employing a black Cu₃VS₄ photocathode for solar CO₂ reduction to produce green syngas. The black metal sulfide was prepared by a solid-state-reaction (SSR) and a flux method using a mixture of chloride salts (flux).¹ The powder-based Cu₃VS₄ photocathode was fabricated by a particle transfer method contacted with deposited Au as a conductive layer.^1,3 A sputtered Au cocatalyst was loaded on the photocathode. Photocathodic properties were measured using a three-electrode system under CO₂ atmosphere. The photocathode was irradiated with visible light and simulated sunlight using a 300 W Xe-lamp attached with a long-pass filter and a solar simulator, respectively. The amounts of evolved H₂, CO and O₂ were determined using an online gas-chromatograph. Single particle layers with Cu₃VS₄ (SSR, flux) particles were observed on the photocathodes by SEM images. Large Cu₃VS₄ (SSR) particles with a size of 5-8 µm contacted with deposited Au layer roughly, while fine contacts with the Au layer were constructed due to small Cu₃VS₄ (flux) particles with a size of 200-800 nm. The results of SEM-EDS mappings and XPS measurements indicated that the surfaces of Cu₃VS₄ (SSR, flux) particles were mostly covered with a sputtered Au cocatalyst and a deposited Au layer. Photoelectrochemical CO₂ reduction to form a green syngas using the Cu₃VS₄ photocathodes proceeded under visible light irradiation. During the reaction, green syngas including 24% of CO was successfully obtained using a Cu₃VS₄ (flux) photocathode. The comparatively high selectivity for CO evolution was due to the 2-type Au functioning as efficient active sites. It was notable that a stability of the photocurrent was drastically improved by employing fine Cu₃VS₄ (flux) particles instead of large Cu₃VS₄ (SSR) particles. XRF analysis and SEM images of the photocathode revealed that a large amount of the Cu₃VS₄ (SSR) particles were peeled off from the Au-deposited substrate after the reaction, while the Cu₃VS₄ (flux) particles were not. Therefore, the good stability of the photocurrent on the Cu₃VS₄ (flux) photocathode was due to the good contacts between the photocatalyst particles and the deposited Au layer. The onset of the IPCE agreed well with that of an absorption edge of Cu₃VS₄, reaching up to 820 nm. In other words, the black photocathode produced a green syngas utilizing a whole range of visible light. Finally, a photoelectrochemical cell consisting of the Cu₃VS₄ photocathode and a BiVO₄ photoanode was constructed. Solar CO₂ reduction proceeded applying the bias at 1.0 V between the photocathode and the photoanode. H₂ and CO steadily evolved accompanied by O₂ evolution from water, reaching unity of the ratio of reacted electrons to holes. These results indicated that solar CO₂ reduction to form a green syngas using the Cu₃VS₄ photocathode was successfully achieved utilizing water as an electron donor. Thus, it was concluded that a Cu₃VS₄ photocathode with the high selectivity for CO evolution and the response to long wavelength visible light was successfully developed. Our findings will contribute to development of a photoelectrochemical cell for an efficient solar energy conversion and a sustainable carbon neutrality. [1] H. Fukai, K. Nagatsuka, Y. Yamaguchi, A. Iwase, A. Kudo, ECS J. Solid State Sci. Technol. 2022, 11, 063002. [2] S. Yoshino, T. Takayama, Y. Yamaguchi, A. Iwase, A. Kudo, Acc. Chem. Res. 2022, 55, 966. [3] T. Minegishi, N. Nishimura. J. Kubota, K. Domen, Chem. Sci. 2013, 4, 1120. Figure 1