Liquid Bismuth Catalyst Enables High‐CO‐Selectivity in CO₂ Hydrogenation

ABSTRACT

Here, we introduce a dynamic liquid‐bismuth (Bi) catalyst that exploits a reversible Bi³⁺/Bi⁰ redox cycle to address the challenge of poor CO selectivity in the reverse water‐gas shift (RWGS) reaction at moderate temperatures. Molten Bi nanodroplets were stabilized via in situ confinement during H₂‐induced reduction of a BiVO₄ precursor, resulting in their firm anchoring on a defective vanadium oxide (VO_x) support while preserving dynamic surface mobility at 400°C, striking a balance between stability and fluidity. Combined mechanistic studies and DFT calculations revealed that Ni‐Bi dual sites on oxygen‐deficient VO_x facilitate an H₂‐assisted redox mechanism: Bi sites enable direct CO₂ dissociation and weak CO^* binding, favoring rapid CO desorption, while the Ni cocatalyst promotes H₂ dissociation and induces electron enrichment of VO_x for CO₂ adsorption. This work demonstrates the potential of liquid‐Bi‐based catalysts for selective and efficient CO₂‐to‐syngas conversion.

This work introduces a dynamic liquid bismuth catalyst for efficient CO₂‐to‐syngas conversion. Stabilized molten Bi nanodroplets on a defective VO_x support enable a reversible Bi³⁺/Bi⁰ redox cycle at moderate temperatures. Mechanistic insights reveal that Ni‐Bi dual sites synergistically activate H₂ and cleave CO₂, leading to high CO selectivity.

1. Introduction

The catalytic hydrogenation of CO₂ via the reverse water‐gas shift (RWGS) reaction represents a pivotal route for CO₂ utilization, producing syngas (CO/H₂) as a feedstock for Fischer–Tropsch and methanol synthesis processes [1, 2]. As an endothermic process, the RWGS reaction is thermodynamically favored at elevated temperatures (500–700°C). However, operating at lower temperatures is desirable for improved energy efficiency. A key challenge arises from the competing, highly exothermic CO₂ methanation reaction, which dominates below approximately 500°C and severely compromises CO selectivity [3, 4, 5]. Therefore, the development of catalysts that kinetically promote CO formation while suppressing over‐hydrogenation to methane is essential for efficient and selective CO₂‐to‐syngas conversion.

Two distinct mechanisms govern RWGS catalysis: the formate pathway and the redox pathway [6, 7, 8, 9]. In contrast to the formate route with multi‐step proton‐coupled electron transfer, the redox pathway facilitates direct C═O bond cleavage, enabling significantly superior CO selectivity. Conventional transition metals (e.g., Cu, Ni, and Pt) typically exhibit strong CO binding, encountering inherent challenges in producing unwanted methane [10, 11, 12, 13]. While strategies such as surface encapsulation or alloying can moderate CO adsorption, they often result in low activity [14]. Bismuth‐based catalysts have recently emerged as promising redox mediators due to their innate resistance to over‐hydrogenation and favorable CO₂ activation capability [15]. Metallic Bi participates in reversible redox cycles (Bi³⁺/Bi⁰), enabling CO₂ reduction with high CO selectivity, as demonstrated in olefin synthesis and related CO₂ hydrogenation processes [15]. However, the low melting point of Bi (275°C) poses a severe constraint: under typical catalytic temperatures (> 300°C), Bi sinters and aggregates, leading to rapid deactivation [16, 17].

Although supported liquid metal systems have been proposed to exploit the dynamic surface of low‐melting‐point metals, achieving a stable and finely dispersed liquid Bi phase remains particularly challenging. This is largely due to Bi's high surface tension and poor wettability on commonly used oxide supports, which often lead to droplet aggregation and catalyst deactivation [18, 19]. Our group recently developed a promising strategy to anchor active metal phases via the exsolution from a solid precursor, which generates nanostructured catalysts with robust metal‐support interactions and enhanced stability [20, 21, 22, 23]. Accordingly, bismuth vanadate (BiVO₄) has attracted attention as a potential precursor, as its reduction under H₂ can produce metallic Bi dispersed on a reducible VO_x support possessing abundant oxygen vacancies [24]. The multivalent nature of vanadium oxide facilitates vacancy formation, which is critical for CO₂ activation. Nonetheless, the dynamic behavior of the Bi/VO_x interface under RWGS conditions, as well as the role of secondary metals in enhancing H₂ activation without compromising CO selectivity, has not been fully elucidated.

In this work, we designed a dynamic liquid‐Bi catalyst stabilized on a defective VO_x support through H₂‐induced exsolution from a BiVO₄ precursor. Although exhibiting high surface mobility due to the low melting point of Bi, the resulting molten Bi nanodroplets remain confined and anchored to the VO_x surface. We further introduce trace Ni (1.6 wt.%) as a H₂ dissociation promoter into the catalyst surface, creating a ternary Ni‐Bi/VO_x catalyst that significantly enhances RWGS performance. Through in situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) and density functional theory (DFT) simulations, we elucidate the H₂‐assisted redox cycle involving Ni‐Bi dual sites on defective VO_x: Bi enables direct CO₂ dissociation and allows for weak CO^* adsorption, while the Ni cocatalyst facilitates H₂ dissociation and induces electron‐rich VO_x for CO₂ enrichment of the catalyst surface. This study demonstrates the potential of liquid‐Bi‐based catalysts for selective CO₂ conversion to syngas, providing a new strategy for designing dynamic catalytic interfaces for renewable fuel and chemical production.

2. Results and Discussion

2.1. Fabrication and Structural Characterization of Bi/VO_x Catalyst

The synthesis of Bi/VO_x catalysts via our “solid precursor” approach is schematically illustrated in Figure S1. Specifically, bismuth vanadate (BVO) precursors were first synthesized hydrothermally using polyvinylpyrrolidone (PVP) as a morphological template. These BVO precursors functioned as sacrificial templates, transforming under reducing reaction atmospheres to generate supported Bi/VO_x catalysts (denoted u‐BVO) via in situ metal exsolution (Figure 1a). Scanning electron microscopy (SEM) reveals that both pristine BVO and u‐BVO retain characteristic rice‐like morphologies (Figure S2a,b). Critically, u‐BVO surfaces display numerous particles confirmed as metallic bismuth by Energy‐dispersive X‐ray spectroscopy (EDS) mapping (Figure S2c).

FIGURE 1.

Microscopic characterization of u‐BVO catalyst. (a) illustration of the structure of the designed Bi/VO_x catalyst. (b) SEM image of u‐BVO (Inset: size statistic of Bi nanoparticle on the catalyst surface). (c) DSC analysis of Bi, BVO, and u‐BVO. (d) TEM image with corresponding elemental mappings of u‐BVO. (e) XRD patterns, and (f) Raman spectra of BVO and u‐BVO. (g) XPS profile of V 2p for u‐BVO. (h) SEM image corresponding to the elemental maps of BVO‐R600. (i) Illustration of liquid Bi nanodroplets migration and coalescence upon cooling.

Two distinct bismuth populations are identified (Figure 1b): surface‐anchored nanoparticles (ca. 58 nm, white arrow) and larger free‐standing particles (> 0.5 µm, yellow arrow). Differential scanning calorimetry (DSC, Figure 1c) reveals Bi's intrinsic melting point (∼275°C), confirming its molten state during catalytic operation (> 300°C). Consequently, the coexistence of these Bi morphologies arises from inherent molten‐Bi migration and coalescence upon cooling (room‐temperature). Importantly, despite macroscopic segregation, nano‐sized Bi particles remain tenaciously anchored, indicative of strong interfacial interactions with the VO_x support. This heterogeneous distribution is further resolved by Transmission electron microscopy (TEM) with correlated EDS elemental mapping (Figure 1d). High‐resolution TEM (HRTEM) imaging (Figure S2d) reveals lattice fringes at 3.28 Å, indexed to the Bi (012) plane. Additional spacings at 3.61 and 3.17 Å correspond to the (102) and (012) planes of Bi₂O₃, indicating partial surface oxidation of Bi domains. Notably, crystalline vanadium oxide phases remain undetected, consistent with an amorphous VO_x support structure as reported previously [24]. Based on ex situ SEM and TEM observations (Figure 1b,d), the liquid Bi catalyst solidifies into discrete particles distributed both on the surface and within the internal regions upon cooling; during the active reaction states, however, these particles would melt into liquid metals anchored by the VO_x support. It should be noted that a dynamic phase transition remains unobservable due to the absence of in situ high‐temperature SEM capabilities in this study.

X‐ray diffraction (XRD) analysis (Figure 1e) confirms the reduction‐induced transformation of BiVO₄ precursors into metallic Bi (JCPDS No. 44–1246) under H₂, while the VO_x support retains an amorphous structure. Raman spectroscopy (Figure 1f) corroborates the presence of VO_x, revealing the characteristic V─O stretching vibration at 812 cm⁻¹. Weak bands at 140 and 193 cm⁻¹ suggest partial oxidation of VO_x to V₂O₅ [25]. Fourier‐transform infrared (FTIR) spectroscopy (Figure S3a) further supports this assignment, displaying characteristic V─O─V (748 and 515 cm⁻¹) and V═O (986 cm⁻¹) stretching vibrations associated with V₂O₅ formation [26, 27]. X‐ray photoelectron spectroscopy (XPS) probed surface chemical states, revealing mixed vanadium valences. Deconvolution of the V 2p_3/2 core level (Figure 1g) demonstrates the coexistence of substantial V⁵⁺ (517.4 eV) and V⁴⁺ (516.5 eV) species [28]. Additionally, the Bi 4f spectrum (Figure S3b) can be deconvolved into signatures for Bi⁰ (157.5 and 162.8 eV) and Bi³⁺ (159.4 and 164.7 eV). The high Bi³⁺/Bi⁰ intensity ratio (ca. 16:1) indicates significant surface oxidation of metallic Bi to Bi₂O₃ upon ambient exposure. Collectively, these results unequivocally demonstrate the in situ formation of liquid Bi nanodroplets dispersed on amorphous VO_x matrices derived from the controlled reduction of the BVO precursor under reduction conditions.

To elucidate the interfacial interactions between Bi and the VO_x support, BVO precursors were subjected to thermal treatment under an H₂ atmosphere at 400°C and 600°C for 3 h, yielding catalysts denoted as BVO‐R400 and BVO‐R600, respectively. SEM and correlated EDS analyses (Figures S4 and S5) reveal complete decomposition of the BVO precursor at 600°C, generating abundant free‐standing Bi particles alongside fragmented VO_x domains. Crucially, partially preserved VO_x adopts a distinctive coral‐like morphology with extensive void structures (Figure S4c). EDS elemental mapping further identifies pronounced Bi enrichment within the core of these coral‐like VO_x aggregates (Figure 1h), confirming effective encapsulation of Bi domains. It should be clarified that Bi, in situ derived from BVO, is restricted to the pore network of the VO_x matrices, rather than being limited solely to the core. These observations demonstrate that the VO_x matrices spatially confine molten Bi, thereby stabilizing surface‐anchored Bi nanodroplets against coalescence and migration during high‐temperature reactions.

Accordingly, we propose a structural evolution pathway for Bi/VO_x formation from the BVO precursor (Figure S6), based on time‐resolved SEM observations (Figure S7) and the above discussion. During H₂ reduction at 400°C, metallic Bi progressively exsolves from the BiVO₄ lattice, nucleating as well‐dispersed nanodroplets anchored across the VO_x surface. These nanodroplets, composed of low‐melting‐point Bi (∼275°C), adopt a molten state under operational conditions (> 300°C). Molten Bi nanodroplets migrate and coalesce on both internal and external VO_x surfaces via a particle migration‐coalescence (PMC) mechanism (Figure 1i) [29]. Crucially, at catalytic temperatures (e.g., 400°C), liquid Bi nanodroplets remain anchored to VO_x yet exhibit dynamic inner and surface mobility, reflecting a balance between confinement and fluidity.

Inspired by the spatial confinement of molten Bi by the VO_x matrices, we hypothesized that precursor morphology modulates interfacial stability. We then synthesized three distinct BVO morphologies: rice‐like particles (BVO‐r, ∼1.9 µm long; Figure S2a), disc‐shaped crystals (BVO‐c, ∼0.7 µm diameter; Figure S8a,c), and sheet‐assembled aggregates (BVO‐a, diameter > 3 µm; Figure S8b,d). Despite morphological differences, all precursors exhibit nearly identical reduction profiles (Figures S9 and S10), confirming that structural evolution is governed by intrinsic chemistry rather than macroscopic form. These morphologically tailored precursors were subsequently used to interrogate structure‐function relationships in CO₂ hydrogenation.

2.2. Fabrication of Ternary Ni‐Bi/VO_x Catalyst

A critical challenge lies in the oxygen incorporation during the redox cycle in the RWGS reaction, which can lead to the formation of surface oxides that degrade the catalyst's structural integrity or cause passivation [30]. To ensure sustained catalytic activity, the kinetics of oxide reduction must dominate over oxidation, a requirement that necessitates a high concentration of surface‐reactive hydrogen species. This can be effectively engineered by creating hydrogen spillover from specific dissociation sites to adjacent active surfaces, thus maintaining the redox cycle and stabilizing catalytic performance [31].

To incorporate active sites for H₂ dissociation, the Ni cocatalyst was anchored onto bismuth vanadate (BVO) precursors via impregnation, followed by in situ reduction to yield the ternary Ni‐Bi/VO_x (denoted u‐Ni/BVO). Inductively coupled plasma‐optical emission spectrometer (ICP‐OES) quantified the elemental composition, revealing a Ni loading of 1.6 wt.%, alongside principal components of Bi (73.4 wt.%) and V (16.7 wt.%) (Table S1). XRD analysis of u‐Ni/BVO (Figure S11a) exhibited two weak diffraction peaks at 2θ = 43.3° and 44.4°, corresponding to the (012) plane of NiO and the (111) plane of metallic Ni, respectively. High‐resolution TEM (HRTEM) images (Figure S11b) further identified lattice fringes with a spacing of 2.10 Å, indexed to the NiO(200) plane. XPS results (Figure S11c) confirmed the coexistence of metallic Ni⁰ (853.0 eV) and Ni²⁺ (855.7 eV) [32]. Moreover, EDS mappings (Figure S12) demonstrated a highly dispersed distribution of Ni species. Collectively, these findings indicate the successful loading of well‐dispersed Ni nanoparticles, which predominantly exist as metallic Ni under H₂‐rich conditions, with the observed NiO phase attributable to surface passivation upon air exposure. Additional characterization via SEM, TEM, XRD, and Raman spectroscopy (Figures S11–S13) consistently confirmed that the introduction of Ni cocatalyst does not perturb the structural evolution of the Bi/VO_x support during reduction.

The role of Ni cocatalyst in modulating the reducibility of BVO was investigated by H₂ temperature‐programmed reduction (H₂‐TPR). As shown in Figure 2a, reduction peaks at 321°C (α‐NiO, weak metal‐support interaction) and 412°C (β‐NiO, moderate interaction) emerge [33, 34]. Critically, Ni incorporation lowers the primary reduction temperature of BiVO₄ from 684°C to 537°C, consistent with H₂ spillover from Ni to BVO [35]. This demonstrates that H₂ dissociation at Ni sites generates active hydrogen species (H^*), which subsequently migrate to the BVO surface, accelerating the reduction of Bi³⁺ (from BiVO₄ precursor) to metallic Bi⁰. Furthermore, compared with u‐BVO, the enhanced HD signal on u‐Ni/BVO during the H‐D isotope exchange experiment provides direct evidence for the crucial role of Ni sites in H₂ dissociation (Figure S14). Theoretically, H₂ adsorption energy increases sequentially (Figure 2b): VO_x (‐0.09 eV) < Bi/VO_x (‐0.13 eV) < Ni‐Bi/VO_x (‐0.32 eV), with corresponding electron transfers of 0.003, 0.018, and 0.027 |e|, respectively (Figure S15). These results establish Ni sites as pivotal centers for H₂ dissociation.

FIGURE 2.

Characterization of u‐Ni/BVO catalysts. (a) H₂‐TPR profiles of BVO and Ni/BVO precursors. (b) The adsorption free energies of CO₂ and H₂ over VO_x, Bi/VO_x, and Ni‐Bi/VO_x. (c) XPS profile of V 2p for u‐Ni/BVO. (d) EPR spectra of u‐BVO and u‐Ni/BVO. (e) The electron localization function of V₂O₅, VO_x, Bi/VO_x, and Ni‐Bi/VO_x. (f) CO₂‐TPD profiles of u‐BVO and u‐Ni/BVO.

XPS analysis (Figure 2c) reveals a 0.2 eV decrease in the V 2p_3/2 binding energy for u‐Ni/BVO (517.2 eV) relative to u‐BVO (517.4 eV), signifying electron enrichment on the VO_x support induced by Ni incorporation. This electronic modulation is corroborated by Raman spectra, where the V─O stretching vibration shifts from 812 to 810 cm⁻¹ (Figure S13) [36, 37]. Deconvolution of the O 1s spectrum (Figure S16) resolves three distinct states: lattice oxygen (O_L), oxygen vacancies (O_V), and surface hydroxyl groups (O_A). Notably, u‐Ni/BVO exhibits a higher oxygen vacancy concentration (18.0%) than u‐BVO (15.0%), consistent with enhanced electron paramagnetic resonance (EPR) intensity at g = 2.001 (Figure 2d). Both O_L and O_V components in u‐Ni/BVO undergo subtle shifts toward lower binding energies (O_L: 530.4 vs. 530.2 eV; O_V: 532.2 vs. 532.1 eV), confirming electron transfer from Ni⁰ to VO_x via interfacial Ni‐O‐V linkages [38].

Electron localization function (ELF) mapping (Figure 2e; Figure S17) visualizes pronounced electron accumulation near interfacial O_V in Ni‐modified catalysts. This localized electron density enhances the electron‐donating capacity of VO_x, a critical feature for CO₂ adsorption. Bader charge analysis quantifies stronger charge transfer (0.041 |e|) from Ni‐Bi/VO_x catalyst to adsorbed CO₂ versus isolated VO_x or Bi/VO_x (Figure S18). Consequently, CO₂ adsorption free energy intensifies from ‐0.15 eV (VO_x) and ‐0.21 eV (Bi/VO_x) to ‐0.40 eV on Ni‐Bi/VO_x, highlighting Ni‐promoted CO₂ adsorption on VO_x (Figure 2b). Experimentally, CO₂ temperature‐programmed desorption (CO₂‐TPD) confirms enhanced CO₂ adsorption capacity in Ni‐containing catalysts (Figure 2f). Thus, the Ni cocatalyst exhibits a bifunctional mechanism: it facilitates H₂ dissociation while concurrently inducing an electron‐rich VO_x surface that promotes CO₂ enrichment on the surface.

2.3. Catalytic Performance for CO₂ Thermal Hydrogenation

The CO₂ thermal hydrogenation performance of u‐BVO was evaluated at 1 MPa and 8000 mL·g_cat ⁻¹·h⁻¹ over 325–450°C (Figure 3a). CO₂ conversion increased with temperature, yielding CO predominantly alongside minor CH₄. The u‐BVO catalyst achieved a CO₂ conversion of 10.0% with a CO selectivity of 79.0% at 400°C. Upon incorporation of Ni at an optimal loading of 1.6 wt.%, the CO₂ conversion increased to 20.0% while the CO selectivity rose to 90.0% (Figure 3b). This enhancement arises from the hydrogen‐rich environment generated by H₂ dissociation on Ni sites, along with improved CO₂ enrichment on the electron‐rich VO_x surface promoted by Ni. The fact that CO₂ conversion showed limited dependence on Ni content, while CO selectivity remained consistently near 90.0% (Figure S19), suggests that the reaction likely follows a direct CO formation pathway. In contrast, a formate‐mediated mechanism would be expected to exhibit stronger sensitivity to H^* availability, which is modulated by Ni concentration during hydrogenation steps. Although the overall activity does not surpass that of leading Ni‐based catalysts (Table S2), these results constitute a significant demonstration of applying a liquid bismuth strategy to thermocatalytic CO₂ hydrogenation while maintaining high CO selectivity.

FIGURE 3.

Catalytic performance for CO₂ hydrogenation. (a) CO₂ conversion and product selectivity of u‐BVO. (b) CO₂ conversion and CO selectivity for u‐BVO and u‐Ni/BVO. (c) CO₂ conversion for the as‐prepared catalysts with different precursor morphologies. (d,e) CO₂ conversion and CO selectivity for (d) Bi‐based or V‐based catalysts, and (e) HCl post‐treated catalysts.

Catalyst morphology modulated activity: sheet‐assembled BVO‐a achieved the highest CO₂ conversion (11.0%) at 400°C, surpassing rice‐like BVO‐r (10.0%) and disc‐like BVO‐c (7.0%) (Figure 3c; Figure S20). Regardless of the catalyst morphology, the product distribution remained largely similar, yielding a CO selectivity of ∼78.0% and a CH₄ selectivity of ∼22.0% at 400°C. Scherrer analysis (Table S3) confirms severe PMC of Bi upon cooling in u‐BVO‐c, yielding larger Bi particles (60.8 nm vs. 47.9 nm for u‐BVO‐r, 47.8 nm for u‐BVO‐a), while the coral‐like VO_x framework in BVO‐a/BVO‐r suppresses Bi sintering through spatial confinement. Ni immobilization further enhanced CO₂ conversion (400°C): Ni/BVO‐c (9.8%) < Ni/BVO‐a (16.2%) < Ni/BVO‐r (20.0%) with comparable CO selectivity (∼90%). Negligible enhancement in Ni/BVO‐c arises from its stacked disc‐like structure, which fails to prevent the coalescence of molten Bi (Table S3). This highlights that ensuring sufficient and accessible Bi active sites is crucial for achieving CO₂ conversion in this system, which can be modulated by support morphology.

Control experiments elucidate critical Bi‐VO_x synergy (Figure 3d). Catalysts lacking integrated Bi/VO_x interaction exhibited low activity (<5.0% CO₂ conversion), with activity trending: Bi/SiO₂+V₂O₃ (physical mixture)> Bi+V₂O₃ (physical mixture)> V₂O₃> Bi/SiO₂> Bi. Negligible activity of Bi (fully reduced to metallic Bi at 400°C; Figures S21 and S22) and sintered Bi particles in BVO‐R600 (4.0% CO₂ conversion) confirm that VO_x‐anchored Bi nanodroplets, not free‐standing Bi, serve as primary active sites. Low activity of Bi dispersed on amorphous SiO₂ from Bi₂SiO₅ (2.0% CO₂ conversion; Figures S21 and S23) and measurable CO₂ conversion on V₂O₃ (2.4% CO₂ conversion; Figures S21 and S22) suggest that VO_x supports provide secondary active sites for CO₂ reduction, though their contribution is subordinate to Bi‐VO_x synergy. These results unequivocally establish that molten Bi nanodroplets anchored on VO_x dominate CO₂ reduction to CO, with Ni optimizing H^* flux and interfacial electron transfer.

To further elucidate the dual role of VO_x in stabilizing Bi nanodroplets and activating CO₂, we subjected u‐BVO to selective surface Bi etching using HCl solutions (concentration: 0.03–0.3 m), yielding catalysts denoted u‐BVO_n (n = HCl concentration; Figures S24 and S25). Increasing HCl concentrations from 0.03 to 0.3 m progressively diminished CO₂ conversion from 4.3% to 0.8% (400°C, 1 MPa), while CO selectivity remained stable (∼75%; Figure 3e). Comprehensive characterization (Raman, SEM, and digital photos in Figures S25–S27) reveals that HCl etching induces two parallel degradation pathways: (i) dissolution of the coral‐like VO_x framework and (ii) accelerated agglomeration of Bi into micron‐scale particles. Crucially, the extent of activity loss directly correlates with the depletion of surface‐anchored Bi nanodroplets, confirming their indispensability for CO₂ activation. These results demonstrate that VO_x provides dual functionality: (i) spatially confining molten Bi to maintain a balance between confinement and fluidity and (ii) enriching CO₂ on its surface to enhance the reaction probability of Bi and CO₂.

Temperature‐swing cycling experiments demonstrated the regenerability of both u‐BVO and u‐Ni/BVO catalysts, attributable to Bi condensation/remelting dynamics (Figure S28). However, over repeated cycles, the optimized u‐Ni/BVO catalyst experienced a slight activity loss, with CO₂ conversion decreasing from 20.0% to 17.0% by the third cycle. To further evaluate operational durability, long‐term stability tests were conducted under standard reaction conditions (Figure S29,). Both catalysts exhibited performance degradation over the 48 h test. Specifically, the u‐BVO catalyst showed a 16.0% activity loss, with CO₂ conversion eventually stabilizing near 8.0%. In contrast, u‐Ni/BVO underwent a more gradual activity reduction of 36.0%, yet its final conversion (12.5%) remained higher than that of u‐BVO, while maintaining CO selectivity around 90.0%. Although the Bi/VO_x phase composition remained unchanged in both catalysts (Figure S30), post‐reaction characterization revealed severe Bi aggregation in u‐Ni/BVO(48 h), as indicated by an increase in the average Bi particle size from 47.0 to 58.2 nm (calculated from XRD; Table S3). In comparison, u‐BVO showed no significant Bi aggregation (45.2 nm vs. 47.9 nm). These findings suggest that catalyst deactivation is primarily driven by the inherent aggregation of Bi. The presence of Ni exacerbates this process, likely by creating an H‐rich surface environment that enhances Bi atom mobility. It has been reported that reductive conditions can weaken metal‐support interactions and promote metal atom mobility, leading to catalyst deactivation [39]. Nonetheless, the relatively stable performance of u‐BVO underscores the importance of balancing Bi confinement and fluidity to sustain catalytic activity.

2.4. Mechanistic Insights Into CO₂ Hydrogenation

In situ DRIFTS was then employed to probe the reaction intermediates over u‐BVO and u‐Ni/BVO catalysts (Figure S31). Catalysts were pre‐reduced in H₂ (400°C, 3 h) and purged with N₂ to clean surfaces. When CO₂ gas was introduced, gaseous CO bands emerged at 2175 and 2115 cm⁻¹ with comparable intensity for both catalysts (Figure 4a), owing to direct CO₂ dissociation. Upon switching the feed from CO₂ to H₂, a distinct peak at 1396 cm⁻¹, assigned to bidentate formate intermediate (HCOO^*), appeared (Figure 4b) [40, 41, 42]. Critically, no decomposition of HCOO^* to CO was observed, ruling out the associative pathway [43, 44, 45, 46]. This confirms CO production predominantly via a redox mechanism, where CO₂ dissociates directly on Bi sites, and H₂ regenerates reduced states of the active site.

FIGURE 4.

Revealing reaction mechanisms of u‐BVO and u‐Ni/BVO catalysts. (a,b) changes in the intensities of IR bands of (a) CO at 2175 cm⁻¹, (b) formate species at 1396 cm⁻¹. (c) CO intensity measured during CO₂ pulse experiments at 400°C. (d) In situ DRIFT spectra of CO desorption at 40°C for u‐BVO and u‐Ni/BVO. (e) in situ DRIFT spectra at 400°C and 1 MPa for u‐BVO and u‐Ni/BVO. (f) H₂‐TPR profiles of u‐BVO, u‐BVO‐CO₂, and u‐Ni/BVO‐CO₂.

CO₂ pulse experiments (Figure 4c) further validate this mechanism: CO₂ pulses immediately generated CO, but subsequent catalyst oxidation suppressed CO yield. Activity was restored after H₂‐mediated re‐reduction, demonstrating that H₂ acts as a reductant for catalyst regeneration, not a direct HCOO^* hydrogenation agent. The intensified CO signal on u‐Ni/BVO (Figure 4a) thus arises from accelerated Bi/VO_x regeneration via H^* spillover from Ni sites, consistent with lower reaction onset temperatures in CO₂/H₂ temperature‐programmed surface reaction (TPSR experiment; Figure S32). Meanwhile, the linear CO^* adsorption at 2064 cm⁻¹ vanished rapidly under N₂ purge (Figure 4d), reflecting weak CO adsorption on u‐BVO and u‐Ni/BVO. This aligned with high CO selectivity, and Ni incorporation did not alter CO adsorption strength. As reported, weak CO adsorption on Bi sites is due to the intrinsic electronic properties of metallic Bi, which lacks a sufficient population of energetically suitable electrons for effective back‐donation into the 2π^* anti‐bonding orbital of CO [47].

High‐pressure in situ DRIFTS (Figure S33) identified additional reaction intermediates. In the case of the u‐BVO catalyst, a linear CO^* peak at 2031 cm⁻¹ (transient intermediate) diminished over time, while HCOO^* (1396 cm⁻¹) and CH₄ (3015 cm⁻¹) intensities increased (Figure 4e), indicating CO^* hydrogenation as the primary CH₄ formation pathway. However, for u‐Ni/BVO, CH₄ was detected only upon switching from CO₂/H₂ to pure H₂ (Figure 4e). This demonstrates that H^*‐mediated rapid regeneration of Bi/VO_x suppresses CO^* accumulation and over‐hydrogenation, favoring CO desorption.

Furthermore, H₂‐TPR quantified the redox cycle dynamics of u‐BVO and u‐Ni/BVO catalysts (Figure 4f). Pre‐treated catalysts (u‐BVO‐CO₂ and u‐Ni/BVO‐CO₂) were exposed to a CO₂ atmosphere after H₂ reduction. Negligible H₂ consumption (5 cm³·g⁻¹) occurred on non‐oxidized u‐BVO, while oxidized u‐BVO‐CO₂ consumed 49.3 cm³·g⁻¹ H₂ (peak at 520°C). These observations strongly support a catalytic redox cycle involving Bi droplets as the primary active sites (i.e., redox cycle, CO₂ + Bi⁰ → CO + Bi₂O₃, H₂ + Bi₂O₃ → H₂O + Bi⁰). In the case of u‐Ni/BVO‐CO₂, H₂ consumption surged to 123.4 cm³·g⁻¹ with a delayed reduction peak (550°C). This arises from a CO₂‐enriched microenvironment around Bi⁰ sites created by Ni‐induced electron‐rich VO_x. Additionally, the Bi₂O₃ overlayer formed during oxidation further stabilizes molten Bi droplets against coalescence [48].

2.5. DFT Calculations

DFT calculations were conducted to unravel the role of O_V in the H₂‐assisted redox mechanism. To emulate the experimental catalyst structures, oxygen‐deficient V₂O₅ (001) surfaces with controlled O_V concentrations were modeled as VO_x, Bi/VO_x, and Ni‐Bi/VO_x systems (Figure 5a; Figure S34). In these models, a static Bi₄ cluster was employed as a targeted computational probe to capture the electronic structure of low‐coordination surface atoms, which represent the most active sites on liquid Bi nanodroplets rather than simulating the entire dynamic droplet. This simplification is justified by the intrinsic atomic‐scale layering and short‐range order characteristic of liquid metal surfaces, which confer transient local rigidity and enable meaningful electronic structure analysis using static cluster models [49, 50, 51]. The formation energy of an additional O_V was calculated to evaluate the ease of O_V generation, following the trend: Bi/VO_x (2.89 eV) < Ni‐Bi/VO_x (3.16 eV) < VO_x (3.24 eV). These results indicate that Bi nanoparticles promote oxygen lattice activation and facilitate O_V formation on the VO_x support. The presence of Ni further fine‐tunes the oxygen vacancy formation energy, reflecting its role in modulating the redox properties of the catalytic interface. This computational approach provides atomic‐level insight into the synergistic roles of Bi and Ni in enhancing the redox activity and stability of the VO_x‐supported catalyst.

FIGURE 5.

DFT calculations. (a) Three catalyst structures modeled VO_x, Bi/VO_x, and Ni‐Bi/VO_x. (b–d) relative energy profiles of different pathways for (b) CO₂ dissociation into CO^*, HCOO^*, and COOH^* at O_V site on Bi/VO_x, (c) direct CO₂ dissociation, (d) H₂ dissociation and regeneration of O_V on VO_x, Bi/VO_x, and Ni‐Bi/VO_x. (e) proposed RWGS reaction mechanism over Bi/VO_x and Ni‐Bi/VO_x catalysts (H₂‐assisted redox cycle).

Reaction free energy profiles for CO₂ reduction pathways are shown in Figure 5b and Figure S35 and S36. The associative pathway (CO₂ ^* + H^* → HCOO^*) on Bi/VO_x requires a high H₂ dissociation barrier (1.43 eV) and a kinetic barrier of 1.23 eV. In contrast, direct CO₂ dissociation at O_V sites (CO₂ ^* → CO^* + O^*) exhibits a lower kinetic barrier (1.12 eV), thermodynamically and kinetically favoring this route over HCOO^* or COOH^*‐mediated paths, consistent with the above‐mentioned in situ DRIFTS data. The dissociated CO^* adsorbs on adjacent Bi sites, but desorbs rapidly due to weakly adsorbed CO on adjacent Bi sites (E _ads = −0.20 eV), aligning with high experimental CO selectivity. Incorporation of Bi or Ni‐Bi further reduces the direct CO₂ dissociation barrier at O_V from 1.05 eV (VO_x) to 0.91 eV (Bi/VO_x) and 0.93 eV (Ni‐Bi/VO_x) (Figure 5c; Figure S37). Although CO binds strongly on Ni sites (E _ads = −1.60 eV), abundant neighboring Bi sites attenuate adsorption strength, ensuring efficient CO desorption after its production.

H₂ dissociation barrier decreases dramatically at Ni sites (0.23 eV) versus Bi (1.43 eV) and VO_x (1.69 eV) (Figure 5d; Figure S38), confirming Ni as the dominant H₂ activation center. Active H^* species migrate from Ni to the Bi/VO_x surface, participating in redox cycles. The regeneration of O_V via H₂O^* formation proceeds most efficiently at Bi─O sites (barrier: 0.67 eV) compared to Ni─O (1.33 eV) and O─O (0.92 eV). Therefore, the synergy of Ni‐Bi facilitates the O_V regeneration on VO_x.

Based on the analysis, we propose a comprehensive H₂‐assisted redox mechanism for the RWGS reaction over the ternary Ni‐Bi/VO_x catalyst, as illustrated in Figure 5e. The mechanism involves two primary steps: (a) CO₂ enrichment and dissociation, where CO₂ is concentrated on the Ni‐induced electron‐rich VO_x surface and undergoes direct dissociation at Bi and O_V sites, consuming metallic Bi and filling vacancies (CO₂ + Bi⁰ → CO + Bi₂O₃; CO₂ + O_V → CO + O^*); and (b) reductive regeneration, where H₂ dissociates at Ni sites to generate mobile H^* species that reduce the oxidized Bi and replenish the oxygen vacancies, producing H₂O (H₂ + Bi₂O₃ → H₂O + Bi⁰; 2H^* + O^*‐O_V → H₂O^* + O_V). Water desorption subsequently regenerates the oxygen vacancies, completing the catalytic cycle. In this scheme, the Ni cocatalyst plays a dual role: (i) it induces electron enrichment of the VO_x support to enhance CO₂ adsorption, and (ii) it provides a continuous supply of H^* for the rapid regeneration of active Bi sites. Concurrently, Bi facilitates weak CO adsorption, enabling high CO selectivity and sustained activity. The high Bi content (73.4 wt.%) and the dense population of Bi nanodroplets on the VO_x surface favor a direct redox pathway. Furthermore, the Ni‐promoted electron‐rich VO_x support creates a CO₂‐enriched microenvironment at the interface, enhancing the interaction between CO₂ and the active Bi/O_V sites (Figure S39).

The molten state of the Bi nanodroplets serves as a central, dual‐function feature that governs both the high efficiency of CO₂ reduction and the catalyst's eventual deactivation. On one hand, the inherent fluidity of the liquid Bi phase enables continuous surface renewal, promoting a highly dynamic Bi³⁺/Bi⁰ redox cycle essential for CO₂ decomposition [18]. This behavior is also observed in other liquid metal systems such as In/In₂O₃ and Ga/Ga₂O₃ [19, 52]. On the other hand, the high surface energy and mobility of the liquid droplets drive particle migration and coalescence (PMC), leading to irreversible agglomeration and consequent loss of active surface area over time.

3. Conclusion

In summary, this work demonstrates a rational strategy for synthesizing molten Bi‐based catalysts that achieve effective thermal CO₂ hydrogenation to CO. Liquid Bi nanodroplets, spontaneously formed and spatially confined on VO_x matrices, serve as robust active sites for CO₂ activation. Multiscale characterization and controlled experiments reveal that metallic Bi dynamically exsolves from BiVO₄ precursors during reaction, forming molten Bi nanodroplets dispersed on amorphous VO_x matrices by strong Bi/VO_x interactions. We establish an exclusive H₂‐assisted redox mechanism for the RWGS reaction: (i) direct CO₂ dissociation occurs at Bi sites and O_V on VO_x; (ii) the incorporation of Ni cocatalyst accelerates Bi and O_V regeneration via H‐species spillover, while simultaneously enhancing CO₂ enrichment on the surface of VO_x. Critically, high CO selectivity stems from two synergistic effects: weak CO adsorption on metallic Bi⁰ sites, and efficient O_V regeneration mediated by Bi─O interfaces. This study serves as an initial investigation into the application of liquid metals in thermocatalytic CO₂ hydrogenation. Although the catalytic performance, particularly in terms of CO₂ conversion, does not yet rival that of state‐of‐the‐art RWGS catalysts operating at high space velocities, it highlights the importance of meticulously engineering the metal‐support interface for developing high‐performance liquid metal catalysts. The viability of supported liquid metal systems hinges on two critical factors: (a) achieving uniform dispersion of liquid metal droplets across the support and (b) ensuring their stability under reaction conditions. Therefore, addressing these interfacial challenges will be the primary focus of our ongoing efforts to optimize liquid Bi‐based catalysts for efficient RWGS conversion.

Conflicts of Interest

The authors declare no conflict of interest.

Supporting information

Supporting File: advs74745‐sup‐0001‐SuppMat.docx.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.