Solar energy driving CO₂ conversion into valuable chemicals is a promising way to tackle the dilemma of climate crisis and energy shortage^[1]. Since the discovery that TiO₂ can be used to reduce H₂O and CO₂, a variety of semiconductor photocatalysts have been fabricated to increase the photo-to-chemical efficiency. Based on these materials, a series of improvement strategies have been developed, such as defect engineering^[2], surface plasmon sensitization^[3], element doping, and heterojunction construction^[4]. Most strategies gain improved photocatalytic performance by increasing visible light absorption^[5], promoting charge separation/transfer and enhancing CO₂ adsorption/activation.

Among the recent reports on metal oxide based photocatalysts, oxygen vacancy (O_v) engineering has been found to be a facile and effective way to improve photoactivity on all respects. Formation of O_v in Bi₂MoO₆ created a defect energy level in the band gap and enhanced the visible-light-driven photocatalytic activity of Bi₂MoO₆^[6]. Our previous work introduced Cu into CeO_2-x to generate and stabilize the surface oxygen vacancies (O_vs) and achieved high CO₂ photoreduction activity^[7]. Though researchers have made great efforts to explore the effects of O_v on catalysts’ chemical/physical properties and photocatalytic reaction process by experiments and theoretical calculations, there are still some issues to be clarified based on rational material design and advanced characterizations, especially in-situ or operando investigations.

Typically, ceria-based oxides serve as an active oxygen donor in a wide range of reactions^[8]. It is worth noting that both Ce³⁺ and Ce⁴⁺ can exist stably because of their unique electronic configurations. To balance the charge, active oxygen is released spontaneously and O_v is formed, which can be usually recovered when exposed to an oxidative environment. Due to this unique nature, great efforts have been made to create O_v and clarify the effects of O_v in CeO₂. In order to demonstrate the effect and behavior of the as-formed O_v in CeO₂, bismuth is selected as a dopant that is expected to maintain the fluorite structure and the stability of O_v because of its lower valence state (+3) and its ion radius (0.117 nm) similar to Ce⁴⁺ (0.097 nm)^[9]. Here we synthesized and explored a Ce_1-xBi_xO_2-δ solid solution photocatalyst with boosted O_v concentration and enhanced CO₂ photoreduction ability under ambient environment. A series of Ce_1-xBi_xO_2-δ oxide catalysts were investigated and the catalyst with an optimal Bi/(Bi+Ce) ratio of 0.4 exhibited highest CO₂ photoreduction efficiency. The relationship between the O_v concentration and the photocatalytic activity of Ce_1-xBi_xO_2-δ solid solution catalyst was explored by structural, compositional and in-situ FT-IR characterizations.

Samples were synthesized by a facile one-step hydrothermal process at controlled conditions. Typically, a certain amount of Ce(NO₃)₃·6H₂O and Bi(NO₃)₃·5H₂O were dissolved in diluted nitrite acid (HNO₃, 5 mL, 4 mol/L) by sonication before the dropwise addition of 8 mol/L aqueous solution of NaOH (35 mL) while stirring. This mixture was kept stirring for 30 min till the formation of a gelatinous suspension and then transferred into a 50 mL Teflon-lined autoclave within a stainless-steel tank and placed at 100 ℃ for 24 h. The obtained samples were rinsed in deionized water and washed with anhydrous ethanol for several times. Then the powder was oven-dried overnight in vacuum at 60 ℃. The samples were denoted as Ce_1-xBi_xO_2-δ, where x represents the molar ratio of Bi/(Ce+Bi). The sample synthesized with Ce(NO₃)₃·5H₂O (1.5 g) as the sole precursor under the same conditions was named as CeO₂.

X-ray diffraction patterns (XRD) of the samples were recorded on Rigaku D/Max 2200PC X-ray diffractometer. The measurement was operated at room temperature under Cu Kα radiation with a scanning rate of 4 (°)/min. A JEM-2100F field emission transmission electron microscope (TEM) (200 kV) was used to obtain the TEM image. Ultraviolet visible (UV-Vis) absorption spectra (800 nm to 200 nm) were recorded by a UV-3101 PC Shimadzu spectroscope (BaSO₄ as the reference standard material). Electron paramagnetic resonance (EPR) spectra were obtained from a Bruker EMXplus (Germany) spectrometer at 90 K. X-ray photoelectron spectra (XPS) were obtained on a Thermo Scientific ESCALAB 250 spectrometer with multichannel detector, Al Kα radiation as excitation source, and C1s at 284.6 eV as a signal- calibrating standard of binding-energy values. Time-resolved fluorescence spectra were tested on an Edinburgh Instruments FLS920 spectrometer under 420 nm excitation of deuterium lamp. DXR Raman microscope was used to obtain Raman spectra under the excitation of 532 nm. In-situ Fourier transform infrared (FT-IR) spectra were collected in N₂ and CO₂, respectively, on Nicolet iS10 equipped with MCT detector. The catalyst was swept by N₂ for 1 h and stabilized for 1 h before the FT-IR spectrum was collected and saved as background data. Subsequently, the catalyst was swept by CO₂ for 1 h and stabilized for 1 h. Then the in-situ FT-IR spectra were recorded at different time intervals since the beginning of light irradiation.

Electrochemical measurements were conducted on a CHI660A electrochemical workstation (Shanghai Chenhua, China) with a standard three-electrode system. Fluorine-doped tin oxide (FTO) glass deposited (15 mm× 25 mm) with photocatalyst as the working electrode, platinum wire and Ag/AgCl as counter electrode and reference electrode, respectively. The electrolyte was 0.2 mol/L Na₂SO₄ aqueous solution. The preparation process of working electrodes by electrophoretic deposition was as follows: sample powder (20 mg) and iodine (10 mg) were mixed and well milled in an agate mortar then well-dispersed in acetone (30 mL). Thus, the plating solution was obtained. A thin film of the sample was uniformly deposited on FTO with a potentiostat bias of 10 V for 10 min and then calcined in an oven at 150 ℃ for 2 h. During the electrochemical measurement, the coated area of all samples was controlled at 1 cm². Mott-Schottky plots were obtained by impedance-potential tests at the frequency of 1000 Hz with a voltage amplitude of 10 mV. Nyquist plots were tested in dark at a bias voltage of -0.4 V vs Ag/AgCl electrode. With irradiation from a 300 W Xe arc lamp, the transient photocurrent densities were obtained at 0.3 V versus Ag/AgCl electrode.

The light source was a 300 W Xenon arc lamp from Aulight CEL-HX, Beijing, the light intensity of which tested on the position of catalyst was 210 mW/cm². The gas products (CO, CH₄) were analyzed by GC-2014 gas chromatograph (flame ionization detector, FID) equipped with a 0.5 nm molecular column and a TDX-01 packed column. The photocatalysis test was conducted typically as follows: 50 mg of photocatalyst was uniformly dispersed on a 2.5 cm × 2.5 cm glass at the bottom of a 625 mL sealed glass reactor. The temperature of the reactor was kept at 15 ℃ with cooling water circulation. To exclude possible influence of contaminants, the sealed reactor was filled with N₂ and subjected to the irradiation of Xe lamp for 2 h before CO₂ (99.99% of purity) moisturized by a water bubbler was pumped into the reactor and stabilized for 30 min. During the photoreaction, 1 mL of sample gas was continually extracted from the reactor every 1 h and analyzed on GC according to external standard method. Before each activity (or cycle) test, the photocatalyst was firstly heat-treated at 150 ℃ for 2 h, so as to remove the organic impurities and adsorbed carbon species.

A series of Ce_1-xBi_xO_2-δ (x=0, 1, 2, 3, 4, 5) samples were synthesized by co-precipitation method and all catalysts were tested for gas-solid phase CO₂ reduction in sealed glass reactor under the conditions of 15 ℃, 1×10⁵ Pa. CO was the main product with trace CH₄ detected. Controlled experiments without light irradiation, CO₂, or photocatalysts were conducted, in which no CO was detected, proving that CO was derived from CO₂ photoreduction. As displayed in Fig. 1, with the increase of Bi content, the catalytic performance of the samples improves. Ce_0.6Bi_0.4O_2-δ shows the highest activity in CO₂ photoreduction with a CO yield of 0.5 μmol·g^-1·h^-1 under Xe light irradiation, which is ~4.6 times of that on CeO₂.

XRD patterns (Fig. S1) show that the Ce_1-xBi_xO_2-δ solid solution samples prepared in this work maintained the fluorite cubic structure of ceria (JCPDS 34-0394). With Bi content ranging from 10% to 50%, all peaks broaden and shift slightly towards small angle, which indicates that adding Bi into ceria increased the lattice parameter. The morphology of the as-prepared ceria and Ce_0.6Bi_0.4O_2-δ were further characterized by TEM (Fig. 2). Pure ceria shows the morphology of granite nanorods, 100-300 nm in length and 12-20 nm in diameter (Fig. 2(a)) while Ce_0.6Bi_0.4O_2-δ is in the form of nanocubes with particle size of ~35 nm (Fig. 2(b)). The substitution of Ce⁴⁺ atom with Bi³⁺ distorted the crystal structure and changed the crystal growth behavior, leading to surface reconstruction and defects, which can be confirmed by the bright spots observed in Ce_0.6Bi_0.4O_2-δ nanocubes^[10]. Fig. 2(c) shows that O, Ce, and Bi distribute homogeneously in the Ce_0.6Bi_0.4O_2-δ sample. No segregation has been observed and a solid solution catalyst can be identified.

To further probe the chemical and compositional differences Bi doping induced, XPS spectra were collected to analyze the chemical states of Ce, Bi and O on the surface of CeO₂ and Ce_0.6Bi_0.4O_2-δ samples. There are multiple peaks in the Ce3d XPS spectra (Fig. 3(a)). Both CeO₂ and Ce_0.6Bi_0.4O_2-δ show eight major characteristic peaks located at 917.9, 916.6, 907.2, 900.7, 898.5, 889.7, 886.5 and 882.3 eV (spin-orbit splitting peaks), which are uniquely attributed to Ce⁴⁺, thus the main valence state of Ce can be identified as +4. The presence of Ce³⁺ can be also distinguished in both CeO₂ and Ce_0.6Bi_0.4O_2-δ by the peaks at 902.5 and 884.2 eV(denoted as v₁ and v₂). Through the comparison of these two characteristic peaks (v₁ and v₂) corresponding to Ce³⁺, especially against the concurrent decrease of the two adjacent peaks(u₁ and u₂) of Ce⁴⁺, we are convinced of that, with the introduction of Bi into ceria lattice, Ce³⁺ increases in Ce_0.6Bi_0.4O_2-δ^[3]. While in Fig. 3(b), the representative core level Bi4f spectrum shows that Bi³⁺ exists exclusively in nanocrystalline solid solution Ce_0.6Bi_0.4O_2-δ^[10]. Two highly symmetric peaks are located at 158.8 eV for Bi 4f_7/2 and 164.1 eV for Bi4f_5/2, which agree well with those observed in standard bismuth oxides^[11]. This result combined with those of XRD and TEM demonstrates that Ce atoms have been successfully substituted by Bi atoms in CeO₂ lattice and a solid solution catalyst has been formed.

The O1s spectra of the samples (Fig. 3(c)) can be deconvoluted into three peaks. For CeO₂, the two peaks at 529.2 and 529.7 eV are assigned to lattice oxygen species (O_latt), O^2- and O₂^2-, respectively^[2]. The peak at 531.6 eV is ascribed to adsorbed oxygen species (O_ad)^[12,13]. It is noted that there is slight band shift of O_ad to lower binding energy in the O1s spectra of Ce_0.6Bi_0.4O_2-δ than in CeO₂ and the signal increases, suggesting the adsorbed H₂O and CO₂ are more easily to get electrons from the surface of Ce_0.6Bi_0.4O_2-δ, that is, are more readily to be activated on the reactive sites after the introduction of O_vs together with Bi. The O_ad content calculated from peak area, which is proportional to that of O_v, is 44%, much higher in Ce_0.6Bi_0.4O_2-δ than 34% in CeO₂. This means our intention of modulating O_v content by Bi-doping is feasible.

In the Raman spectra (Fig. 3(d)), CeO₂ and Ce_0.6Bi_0.4O_2-δ exhibit distinct vibrational bands at 350-650 cm^-1. For pristine CeO₂, the band at 455 cm^-1 is ascribed to the F_2g vibration mode of CeO₂ fluorite while the weak signal at about 600 cm^-1 is related to the intrinsic O_vs due to the existence of Ce^3+[14]. Compared with pure CeO₂, major differences appear in the Raman vibration modes of Ce_0.6Bi_0.4O_2-δ in that the F_2g band shifts to low frequency (about 450 cm^-1) and slightly broadens. This phenomenon is explained as that the symmetrical stretching vibration of Ce[O]₈ unit is distorted by Bi doping. The second vibration mode of Ce_0.6Bi_0.4O_2-δ located at 525 cm^-1 can be indexed to the presence of Bi³⁺. More importantly, the signal around 585 cm^-1, which is related to extrinsic O_vs that serve as charge-compensating defects, is significantly strengthened^[15,16,17]. Here, it can be concluded that Bi doping in CeO₂ has boosted the presence of O_vs. For Ce_0.6Bi_0.4O_2-δ, the half-width at half-maximum (HWHM) at 450 cm^-1 is ~29.8 cm^-1, based on which the intrinsic O_vs concentration is calculated to be ~2.34×10²¹ cm^-3. It is higher than that of the pristine CeO₂ (2.10×10²¹ cm^-3). This can be also verified by the enhancement of EPR signal at g-value of 2.003 corresponding to O_vs of Ce_1-xBi_xO_2-δ compared with that of CeO₂ in Fig. S2. In following discussions, corresponding implications on the catalytic activity of the solid solution catalyst will be illustrated from aspects including visible light absorption, electrochemical performance and CO₂ adsorption/activation.

UV-Vis diffuse absorbance spectra of all the samples were recorded as displayed in Fig. S3. Compared with pure ceria, all Ce_1-xBi_xO_2-δ solid solution samples show enhanced and red-shifted absorption. Among them, samples Ce_0.6Bi_0.4O_2-δ and Ce_0.5Bi_0.5O_2-δ show the strongest UV-visible light absorption that extends from 380 nm to around 500 nm. The result and the yellow color of the solid solution catalyst (pure ceria is white) indicate that the introduction of Bi can successfully extend the light response to UV-visible region, which can be attributed to the defect energy level generated from the increased O_v concentration. Based on the UV-Vis diffuse absorbance spectra and Kubelka-Munk equation^[18] for indirect gap semiconductors, the bandgaps of pure ceria and Ce_0.6Bi_0.4O_2-δ were estimated to be 3.1 and 2.6 eV, respectively. This bandgap narrowing caused by Bi incorporation and O_v can initiate the effective visible light induced photocatalytic CO₂ reduction reaction.

To better understand the influence of Bi doping on the energy band structure of the solid solution catalyst, the Mott-Schottky plots of CeO₂ and Ce_0.6Bi_0.4O_2-δ electrodes were analyzed, both of which are S-type curves of n-type semiconductors (Fig. S4(a)). The flat band potentials (V_fb) were estimated from the linear part of the plots as -0.59 and -0.48 V vs Ag/AgCl electrode for CeO₂ and Ce_0.6Bi_0.4O_2-δ, respectively. According to literature, the conduction band potential (E_cb) of one n-type semiconductor is 0-0.2 V more negative than its V_fb, depending on carrier concentration and effective mass of electron. Here, the potential difference is set as 0.1 V. Based on this result, the band structure can be reasonably estimated in which a downshift of E_cb (from -0.7 V of CeO₂ to -0.6 V of Ce_0.6Bi_0.4O_2-δ) and an upshift of E_vb (from 2.4 V of CeO₂ to 2 V of Ce_0.6Bi_0.4O_2-δ) are clearly illustrated. This band narrowing happens when the Bi6p and Bi6s orbits participate in the construction of CB and VB, respectively^[12].

EIS Nyquist plots of the as-obtained samples in dark demonstrate that Ce_0.6Bi_0.4O_2-δ is much more conductive than CeO₂ (Fig. S4(b)). This is because that ceria is a kind of good ion conductive solid electrolyte and O_v can promote ion transportation^[13] and thus endows the Ce_0.6Bi_0.4O_2-δ solid solution catalyst better conductivity which could efficiently promote the separation and interfacial transmission of photoexcited carriers. As a result, the photocurrent response of Ce_0.6Bi_0.4O_2-δ nearly triples over that of CeO₂ (Fig. S5).

Based on the electronic structure estimation, a possible CO₂ photoreduction mechanism is proposed in Fig. 5. To further understand the origin of enhanced CO₂ photocatalytic reduction activity of the solid solution catalyst, in-situ FT-IR spectroscopy was used to investigate the adsorption, activation, and conversion process of CO₂ on Ce_0.6Bi_0.4O_2-δ and CeO₂. The in-situ FT-IR spectra of CO₂ adsorption on CeO₂ and Ce_0.6Bi_0.4O_2-δ swept by humid CO₂ and stabilized for 1 h were collected and shown in Fig. 4(a, b), respectively. Similar modes of carbonate signals after humid CO₂ sweeping can be observed on the surface of the catalysts, with those of Ce_0.6Bi_0.4O_2-δ more intense^[14]. Specifically, the band at ~1700 cm^-1 corresponds to bicarbonate (HCO₃^-)^[15]. The bands at about 1510 and 1380 cm^-1 are assigned to mono-dentate carbonate (m-CO₃^2-)^[16]. The bands at about 1337 and 1422 cm^-1 correspond to bidentate carbonate (b-CO₃^2-)^[17]. The peaks at about 1255, 1610 and 1650 cm^-1 are assigned to the bending vibration of carboxylate (COO^-)^[18] while the bands at about 1550 and 1200 cm^-1 are ascribed to *COOH. The existence of -CO₃^2- is due to the formation of carbonic acid and the adsorbed CO₂ on the catalyst surface^[2].

After Xe light irradiation, the in-situ FT-IR spectra of CeO₂ and Ce_0.6Bi_0.4O_2-δ were collected and shown in Fig. 4(c, d), respectively. Over irradiation time, the clear increase and decrease of signals that is corresponding to the generation and consumption of the adsorbed intermediates on the surface of Ce_0.6Bi_0.4O_2-δ emerges while those on CeO₂ remain indiscernible. The aforementioned IR bands of both adsorbates and intermediates emerging between 1650 and 1200 cm^-1 show obvious increase with prolonged irradiation time. These observations can be ascribed to the fact that the O_vs could stabilize the carbonate radicals and increase their concentration on the catalyst surface by promoting charge separation and transfer to the adsorbed intermediates^[19,20]. According to literature^[21], the rate limiting step of CO₂ conversion to CO is the conversion of *COOH to *CO, which means the stabilized intermediates, including *COOH, cannot be readily converted to final products and explains the increase of enhanced signal of intermediate carbonates. In contrast, the peak at ~1700 cm^-1 corresponding to HCO₃^- remains unchanged under irradiation. This indicates that bicarbonate plays a different role from other adsorbates that have signals between 1650 and 1200 cm^-1. Recent reports^[22,23] using operando ATR-SEIRAS discussed the effects of bicarbonate in liquid-solid interface reduction of CO₂, that is, the majority of CO₂ reactant near the catalyst surface originates from an equilibrium reaction with HCO₃^- rather than direct adsorption from the solution. Although there is still controversy in the process of CO₂ reduction^[24], here based on our in-situ FT-IR results, it is reasonable to propose that in a gas phase reduction of CO₂ in the presence of H₂O vapor, CO₂ and surface adsorbed H₂O form HCO₃^- and release active species (denoted as init^- in Eq. 1) participating in CO₂ reduction reaction^[25]. The surface concentration of HCO₃^- in the equilibrium does not increase because CO₂ pressure can be seen as constant. This is consistent with the photocatalytic activity test in which Ce_0.6Bi_0.4O_2-δ shows much higher photocatalytic performance than CeO₂ (Fig. 1).

CO₂ (gas)+H₂O_ad « HCO₃^-_ad+H⁺« init^- (Eq. 1)

With the characteristic in-situ FT-IR bands and time related changes under irradiation, we proposed a pre-adsorption process that may shed new light into the mechanism of CO₂ photoreduction (Fig. 5)^[26]. It is also evidenced that the introduction of Bi has created more O_vs and formed localized electrons, which show significant benefit on the adsorption/activation behavior of CO₂ on the surface of the solid solution catalyst and promote the CO₂ photocatalytic activity of the catalyst.

In this work, Bi³⁺ with a lower valence and similar ion radius compared to Ce⁴⁺ has been doped into ceria forming a solid solution photocatalyst, from which it is concluded that:

1) Bi³⁺ doping successfully introduced O_vs into ceria while maintained its fluorite structure, thus the simultaneous introduction of defects as recombination sites of charge carriers could be effectively avoided;

2) The solid solution catalyst showed obviously improved CO₂ photocatalytic reduction activity with a CO yield ~4.6 times of that on pure ceria;

3) The favorable effects of O_vs and surface localized electrons on the adsorption/activation behavior of CO₂ on the catalyst surface and the separation/transfer of photoexcited carriers have been disclosed;

4) Doping aliovalent heteroatoms into metal oxide semiconductors to form solid solution catalysts is an effective way to introduce O_vs and promote photocatalytic activity.

Supporting materials:

Bi-doped Ceria with Increased Oxygen Vacancy for Enhanced CO₂ Photoreduction Performance

LIU Yaxin^1,2, WANG Min^1,2, SHEN Meng^1,2, WANG Qiang^1,2, ZHANG Lingxia^1,2

1. State Key Laboratory of High Performance Ceramics and Superfine Microstructure, Shanghai Institute of Ceramics, Chinese Academy of Sciences, Shanghai 200050, China; 2. Center of Materials Science and Optoelectronics Engineering, University of Chinese Academy of Sciences, Beijing 100049, China