Catalysts are inherently dynamic during a reaction since they could change their local structure for responding to the environment [22–25]. While the reconstruction of catalyst during the reaction is a recognized phenomenon, it can be disclosed via catalytic activity trends. Analyzing the state of reconstruction allows distinguishing real working catalytic sites, so as to design new catalysts [26–28]. In this work, the dynamic surface reconstruction of BOB samples during photocatalytic CO₂ reduction and its influence on the catalytic reaction was studied. The x-ray diffraction pattern matches the tetragonal phase BOB (JCPDS 09-0393) in Fig. S1. The transmission electron microscopy (TEM) image showed 2D sheet-like morphology in Fig. S2A, while the high-resolution transmission electron microscopy (HRTEM) image in Fig. S2B revealed the high orientation along the [010] projection of tetragonal phase BOB, which has an open channel structure with the surface atomic alternation of [Bi₂O₂]²⁺ and Br⁻ on the facet. More importantly, the g value of 2.003 in the EPR spectrum consists of a single line of typical Lorentzian shape, confirming the existence of vacancies after Xe light irradiation (Fig. 1A) [29]. To investigate the types of vacancy generated on the BOB surface after light irradiation, the precipitation of Br⁻ on the surface of BOB samples with different illumination times was determined by ion chromatography (IC). With the prolonged illumination time, the concentration of Br⁻ gradually increased and reached equilibrium after 30 min in Fig. 1B. Therefore, we speculate that Br vacancies are generated on the BOB surface after irradiation. In addition, according to the following formula based on Hooke’s law: =\frac{1}{2\mathit{\pi }c}\sqrt{\frac{k}{\mu }}, where ω is Raman shift (cm⁻¹), c is the velocity of light, and μ is the effective mass. The Raman shift of the Bi–Br bond (e.g., 162 cm⁻¹) reduces after illumination, attributed to the decreased force constant (k) of the Bi–Br bond due to the introduced Br vacancy [30–32]. This result further confirms the generation of Br vacancies during the light irradiation in Fig. 1C.

For surface structure analysis, quasi-in situ XPS was conducted. In the high-resolution XPS Br 3d spectra of the BOB sample in Fig. 1D, there are 2 peaks at 68.38 (Br 3d_3/2) and 68.34 eV (Br 3d_5/2) [33]. After light irradiation of the surface of the BOB sample, the Br 3d shifted to lower binding energy, demonstrating the local chemical environment change resulting from the generation of PI-BrVs. The Bi 4f XPS spectra of BOB exhibit the characteristic spin-orbit doublet splitting center at 159.31 and 164.61 eV in Fig. 1E [34]. In contrast, the Bi 4f spectrum of BOB-light shows a shift toward lower binding energy relative to the BOB, indicating the potential breakage of Bi–Br bonds for PI-BrVs formation. Besides, 3 oxygen-containing species in the O 1s XPS spectra can be assigned to the lattice oxygen atoms near the surface (about 530.14 eV), surface oxygen (about 532.05), and adsorbed oxygen on the sample surface (about 533.32) in Fig. 1F [35,36]. After light irradiation, the O 1s XPS spectra have a slight shift toward lower binding energy, which is attributed to the introduction of Br vacancies. Consequently, Br vacancies are generated on the surface of BOB during light irradiation.

To investigate the electronic interactions between the PI-BrVs and CO₂ molecules, in situ EPR experiments were conducted under a CO₂ atmosphere. The in situ EPR visualization map of the BOB sample is shown in Fig. 2A. The BOB sample was illuminated for 2 h in a vacuum by a 300-W xenon lamp with an AM 1.5 G filter. The light-exposed BOB sample was transferred to a home-built in situ EPR catalytic reactor (Fig. S3). This in situ EPR device can continuously record the spectra under light irradiation, in gas atmosphere, or in a vacuum condition. Then, we recorded the in situ EPR spectra of the light-exposed BOB sample in a vacuum. Under light irradiation for 40 min, the EPR spectra have barely changed, indicating that the surface PI-BrVs have reached saturation after light irradiation of 2 h. Subsequently, EPR spectra intensity reduced with the injection of CO₂ molecules, demonstrating that electrons transfer from the surface of BOB to CO₂ molecules, which is a crucial step for CO₂ molecule activation. Furthermore, theoretical calculation suggests that the adsorbed CO₂ molecules on the BOB sample with Br vacancy have more obvious bending of the molecular structure than the BOB sample (Fig. S4A and B). The non-linear CO₂ molecules are more unstable than linear CO₂ molecules and exhibit a higher reactivity in CO₂ photoreduction. The energy level of the LUMO is lowered with the O=C=O bond angle reducing, thus decreasing the energy barrier for CO₂ activation (Fig. 2C). Therefore, the bent CO₂ molecules could be easily activated via electron transfer from Br vacancy to the unpopulated antibonding 1π_u* orbital of a CO₂ molecule (Fig. S5). Of note, after the injection of CO₂ molecules, the EPR signal linewidth is wider as shown in Fig. 2B. According to Heisenberg’s uncertainty principle, ∆E ∙ ∆t ≈ ℏ can be described by the following equation:

Of note, Δν(=γ_e∆B) (Hz) or ∆B (mT) means the linewidth of the EPR signal, and ∆t (s) stands for the lifetime of the spin state. A long- (or short-) lifetime state would produce a narrow (or wide) EPR signal.

The spin state lifetime (∆t) for α (Ms =  + 1/2; spin up) or β (Ms =  −1/2; spin down) is decided by relaxation times T_1e and T_2e:

T_1e stands for the spin-lattice relaxation (SLR) time and T_2e stands for the spin–spin relaxation (SSR) time. SLR controls the energy exchange between the spin ensemble and the environment (lattice), while SSR includes the interactions within the ensemble. In general, the relationships T_1e ≫ T_2e and 1/T_1e ≪ 1/T_2e generally hold, leading to

Therefore, following the uncertainty principle, the linewidth is

The EPR signal is typically expressed as absorption of the first derivative of A with respect to B as a function of B, dA/dB (Fig. 2E). The shape of A could be a Gaussian curve or a Lorentzian curve or be a suitable mixture of the two, where T_2e is multiplied by a function of T²_1e, with T²_2e either in the exponent (Gaussian) or in the denominator (Lorentzian). The eigen values are the maximum A_max of A, the peak width of its half-height (A_max/2) is ∆B_1/2, and the peak-to-peak distance of the distance ∆B_pp of the derivative curve dA/dB [37,38]. Therefore, the SSR time T_2e decreases as linewidth broadens. Advanced in situ continuous wave-electron paramagnetic resonance (cw-EPR) and pulsed EPR technologies were applied to investigate the interactions between in situ generation Br vacancies and CO₂ molecules during the photoreduction process and reveal the mechanistic insights into the photocatalytic reduction of CO₂, which provides technical support for the study of photocatalytic CO₂ reduction reaction from a fresh point.

In addition, the pulsed EPR measurement of the SSR time T_2e was performed on the X-band EPR spectrometer (CIQTEK, EPR-100) with a dry cooling system. The T₂ relaxation time before and after lighting in different atmospheres (Fig. 2D and Table S1) was compared. In the CO₂ atmosphere, the T₂ relaxation time after light irradiation is reduced by 198 ns compared with that in the dark condition, while it is only reduced by 23 ns in the Ar atmosphere. This result further confirms that CO₂ molecules can lead to the reduction of T₂ relaxation time. Of note, T₂ relaxation time represents the recovery process of the lateral component, which is called lateral relaxation time, also known as SSR time. This process is completed through the energy relay between spin centers as shown in Fig. 3A. Under the Ar atmosphere, a low-energy electron (|β> state) could absorb microwave energy hν and transition to a high-energy state (|α> state), which could transfer the energy to the next low-energy electron and then itself returns to the initial low-energy state (|β> state). Under a CO₂ atmosphere, the excited high-energy electron (|α> state) transfers energy to the CO₂ molecule adsorbed on the surface; high-energy CO₂ molecules are more likely to be activated. The result is further verified by density functional theory calculation; the CO₂ molecule adsorbed on the Br vacancy displays more negative adsorption energy and more pronounced bond angle alteration than that on the BOB sample without Br vacancy, which indicates higher thermodynamics feasibility. Consequently, the EPR linewidth broadens and the relaxation time decreases due to the strong interaction between the spin center and CO₂ molecules, which is ascribed to the in situ generated surface Br vacancies. Subsequently, most of the CO₂ molecules were desorbed from the surface and the EPR signal recovered after vacuum treatment of the sample. Very few CO₂ molecules are strongly adsorbed or undergo chemical reactions, resulting in a small number of single electrons being consumed (Fig. S6). To further explore the interaction between CO₂ molecules and defective surface, the formation of reaction intermediates is monitored by in situ DRIFTS technology in Fig. 3B and C. When only CO₂ molecules flowed into the reactor, the IR peak at 1,421 cm⁻¹ corresponding to ·{\text{CO}}_{2^{-}} can be observed, indicating that adsorbed CO₂ molecules acquire electrons from the PI-BrVs. Additionally, the peak at 1,598 cm⁻¹ and 1,396 cm⁻¹ can be found, which could have contributed to the bidentate carbonate \left(\mathrm{b}-{\text{CO}}_{3^{2-}}\right) group and monodentate carbonate \left(\mathrm{m}-{\text{CO}}_{3^{2-}}\right) group, respectively [39]. This carbonate most likely derived from the interaction between CO₂ and residual water vapor in the reactor, while the peak at 1,598 cm⁻¹ disappeared under light irradiation, demonstrating that the dissociation of the bidentate carbonate occurs in the absence of water vapor. Subsequently, both CO₂ molecules and water vapor were simultaneously introduced into the reaction chamber. In addition, a new IR peak at 1,637 cm⁻¹ was detected, which could be attributed to the COOH* species, considered as the critical active intermediate during photocatalytic CO₂ reduction [40]. Meanwhile, the distinct peaks at 1,608 cm⁻¹ and 3,100 to 3,750 cm⁻¹ can be detected, which could be assigned to the adsorption of H₂O molecules [34]. This result further evidences that the in situ generated PI-BrVs act as active sites to promote the adsorption and activation of CO₂ molecules and H₂O molecules.

Reactant molecule adsorption and activation are an essential step in catalytic reaction, with the H₂O molecule playing a critical role in photocatalytic CO₂ reduction. In general, H₂O molecules can provide a proton source for the CO₂ hydrogenation process. To explore the adsorption and activation of H₂O molecules on the PI-BrVs, an in situ EPR experiment in a moist Ar atmosphere (H₂O/Ar) was carried out. As shown in Fig. 4A, after the generation of PI-BrVs under light irradiation, the EPR signal decreases with the injection of water vapor, indicating that H₂O molecules adsorb on the PI-BrVs and interact with surface single electrons. Subsequently, the EPR signal is partially restored with the removal of H₂O vapor in a vacuum (Fig. S7). This phenomenon demonstrates that, in a vacuum, the weakly adsorbed H₂O molecules desorbed, while the strongly adsorbed or chemically reacted H₂O molecules consumed electrons, leading to the part of the restoration of the EPR signal. Therefore, the H₂O molecule could be adsorbed and activated by electron transfer on PI-BrV sites. The H₂O molecule, as a proton donor during the reduction process of CO₂, consumes electrons to generate the H* and OH⁻.

Importantly, there is no significant change in the linewidth of the EPR spectrum after the injection of H₂O molecules (Fig. 4B), which was different from the phenomenon of direct injection of CO₂ molecules, further indicating the special energy exchange between CO₂ molecules and the surface of PI-BrVs. To study H₂O molecules’ adsorption and activation on the defect surface, in situ DRIFTS was performed in Fig. 4C and D. Upon introducing H₂O molecules into the reaction chamber, an obvious IR peak at 1,616 cm⁻¹ emerged, which could be attributed to the deformation vibrations of adsorbed H₂O molecules. In the meantime, the stretching vibrations of hydroxyl groups (3,050 to 3,700 cm⁻¹) were observed and the peak strength evidently increased with the accumulation of adsorption time. These results suggest that the generation of Br vacancies greatly enhanced the adsorption of H₂O molecules, which lays the foundation for further photocatalytic CO₂ reduction reactions. Subsequently, new peaks at 1,608 cm⁻¹, 1,421 cm⁻¹, and 1,396 cm⁻¹ could be found when CO₂ molecules and H₂O molecules were injected into the reaction chamber together, which could be ascribed to \mathrm{b}-{\text{CO}}_{3^{2-}}, ·{\text{CO}}_{2^{-}}, and \mathrm{m}-{\text{CO}}_{3^{2-}}. More importantly, peaks at 1,635 cm⁻¹ and 1,540 cm⁻¹ were observed and corresponded to COOH*, an essential intermediate in CO₂ reduction to CO. After the light irradiation, the intensity of the peaks (at 1,616 cm⁻¹ and 3,050 to 3,700 cm⁻¹) associated with H₂O molecules decreased dramatically, further evidencing the activation and consumption of H₂O molecules, which could provide efficient proton for further photocatalytic CO₂ reduction reaction.

In the above experiments, the adsorption and activation of CO₂ and H₂O molecules on the PI-BrVs were investigated. To unravel the inherent reason for the CO₂ photoreduction properties, in situ EPR experiments in a mixture atmosphere (CO₂/H₂O) were further performed to explore the real reaction mechanism by simulating an actual photocatalytic environment. When the reactant molecules (CO₂/H₂O) were introduced into the reactor, the EPR signal reduction is attributed to the single-electron transfer from PI-BrVs to CO₂ and H₂O molecules (Fig. 5A). The result indicates that the strong electronic interaction between Br vacancies and reactant molecules could facilitate CO₂ and H₂O molecules’ adsorption and activation. Subsequently, when the reactor is in a vacuum, some of the weakly adsorbed reactant molecules are desorbed from the defect sites, while the strongly adsorbed or chemically reactive molecules cannot desorb from the catalytic surface, resulting in the recovery of partial single-electron signals (Fig. S8). More importantly, the EPR signal line width broadened with the injection of the mixture (CO₂ and H₂O), resulting in a shortened relaxation time. The shorter the relaxation time, the more efficient the energy transfer between spin and environmental, indicating that the reactant molecules interact more strongly with the PI-BrVs. In situ DRIFTS was used to probe reaction intermediates and elucidate the potential reaction pathways (Fig. 5D and E). A new peak at 1,419 cm⁻¹ was observed, corresponding closely to the carboxylate (·{\text{CO}}_{2^{-}}) vibration frequency. Furthermore, CO₂ molecules and H₂O molecules co-adsorbed on the Br vacancies, leading to the creation of the monodentate carbonate \left(\mathrm{m}-{\text{CO}}_{3^{2-}}\right) group at 1,396 cm⁻¹ and the asymmetric OCO stretches of the bidentate carbonate \left(\mathrm{b}-{\text{CO}}_{3^{2-}}\right) group at 1,340 cm⁻¹ [41]. More importantly, the COOH* is regarded as one of the most crucial intermediates for converting CO₂ to CO, and its related peaks at 1,635 cm⁻¹ and 1,540 cm⁻¹ could be found [36,42]. The most likely reduction pathway for photocatalytic CO₂ reduction is proposed as follows based on the above in situ DRIFTS and in situ EPR experiments:

During the process of photoreduction of CO₂, the BOB sample could produce 45.51 μmol g⁻¹ of CO after 5 h of the photoreduction reaction (Fig. 5C). To assess its stability, the BOB sample was tested for an extended period with 10 h of light exposure. Figure S9 shows the linear growth of CO yields with increasing illumination time, providing strong evidence of good stability. In addition, the isotope probing method is employed to trace the source of C, where ¹³CO₂ is used as the carbon feedstock (Fig. 5F). A peak at m/z = 29 was detected by mass spectroscopy, which was ascribed to the ¹³CO. This is powerful evidence for CO generation through light-induced CO₂. In addition, the absence of a ¹³CH₄ peak in the isotopic labeling test is consistent with the absence of CH₄ or H₂ detected in Figs. S10 and S11. This result suggests 100% catalytic selectivity of stable photocatalytic CO₂ reduction reaction on the BOB sample. Compared with previously reported bismuth-based photocatalysts, BOB with PI-BrVs has a higher CO generation rate and product selectivity (Table S2). Overall, the interaction between PI-BrVs and reactant molecules promotes the photocatalytic CO₂ reduction reaction.

All chemicals used were of analytical purity and required no additional treatment.

The BiOBr sample was prepared by a method of precipitation. Details of the experiment are shown in the Supplementary Materials.

The self-made in situ EPR reaction chamber is a quartz sample tube with an inner diameter of 8 mm. The type of EPR spectrometer is EMX nano (Bruker). Details of the experiment are shown in the Supplementary Materials.

Photocatalyst (20 mg) was used in this test. The light source is a 300-W xenon lamp (PLS-SXE300UV, Beijing Perfectlight). The content of Br⁻ was detected by an IC (Shimadzu Essentia LC-16i chromatograph). Details of the experiment are shown in the Supplementary Materials.

In situ DRIFTS test was implemented using a TENSOR II FT-IR spectrometer (Bruker) equipped with an in situ diffuse-reflectance chamber (Harrick) and a high-temperature reaction chamber (HVC). Further experiment details are available in the Supplementary Materials.