Engineered disorder in CO2 photocatalysis

Light harvesting, separation of charge carriers, and surface reactions are three fundamental steps that are essential for an efficient photocatalyst. Here we show that these steps in the TiO2 can be boosted simultaneously by disorder engineering. A solid-state reduction reaction between sodium and TiO2 forms a core-shell c-TiO2@a-TiO2-x(OH)y heterostructure, comprised of HO-Ti-[O]-Ti surface frustrated Lewis pairs (SFLPs) embedded in an amorphous shell surrounding a crystalline core, which enables a new genre of chemical reactivity. Specifically, these SFLPs heterolytically dissociate dihydrogen at room temperature to form charge-balancing protonated hydroxyl groups and hydrides at unsaturated titanium surface sites, which display high reactivity towards CO2 reduction. This crystalline-amorphous heterostructure also boosts light absorption, charge carrier separation and transfer to SFLPs, while prolonged carrier lifetimes and photothermal heat generation further enhance reactivity. The collective results of this study motivate a general approach for catalytically generating sustainable chemicals and fuels through engineered disorder in heterogeneous CO2 photocatalysts.


Point-by-point response
Reviewer #1 (Remarks to the Author): This manuscript presents the beneficial synergy between structural disorder and surface frustrated Lewis pairs in photocatalytic CO2 hydrogenation, with excellent CO production rate and TOF. It is helpful for designing SFLPs-base photocatalysts. However, this manuscript does not bring any new knowledge on materials property and therefore only contribution may be in novel preparation method. The mechanisms of CO2 photocatalytic reactivity and core-shell heterostructure are not very clear; there are several statements not supported with data and even some results cannot distinguish between high precision data and errors. For these reasons, I suggest to reject this paper in the present form for publication in Nature Communications.
Reply: We appreciate the referee's positive comments on our novel preparation method to obtain the black titania. We would like to emphasize that the obtained lack titania showed superior performance towards photo(thermal) CO2-to-CO catalysis, which can compete with reported photothermal catalysts consisting of metal oxides supported transitional metals (Supplementary Table 4). Furthermore, we provide the first molecular-level understanding-the SFLP-to the amorphous black titania during the photo(thermo)catalysis, which offers an understanding advance in the field. During this revision, we added additional experimental results (characterizations and control activity tests), calculations on light penetration depth and TDOS, explanatory sentences and corrected typos to resolve concerns related to the mechanism. Actions taken are indicated in blue while explanatory sentences are shown in red.
Detailed comments are as follows: (1) The authors proposed "UV can penetrate the a-TiO2-x(OH)y thick layer to reach and excite the c-TiO2 component, generating electron-hole pairs", but from TDOS and PDOS relults, the bandgap of amorphous a-TiO2-x(OH)y sheel is smaller than crystalline c-TiO2 core, why the photo-induced electron-hole pairs generate directly in shell layer? That is, why the UV can excite the a-TiO2-x(OH)y component?
Reply: Thanks for the referee's comment, however, we cannot see any conflicts on this point. Theoretically, the smaller the bandgap of a semiconductor, the easier it is to be excited by light. In our case, the UV light can excite both the core c-TiO2 and the shell a-TiO2-x(OH)y because the photon energy of the UV light (λ < 347 nm or E > 3.57 eV) is larger than the bandgap energy of c-TiO2 (Eg = 3.57 eV) or a-TiO2-x(OH)y (Eg = 3.38 eV; Supplementary Fig. 9). Accordingly, the visible and IR light can also excite electrons on the defect-and OH-related energy levels of the shell a-TiO2-x(OH)y, which is shown in Fig. 4d (below picture).
To further support above claims, the penetration depth (Dp) of UV light can be calculated for TiO2: The Dp = 1/, where  is the light absorption coefficient that can be calculated using the formula of  = 4πk/λ, where k is the extinction coefficient that is documented for white TiO2 1 , λ is the wavelength of the incident light. Based on above calculation, the wavelength-dependent Dp is listed below for white TiO2. According to the reflection spectra (Fig. 4a), c-TiO2 and a-TiO2-x(OH)y demonstrates close UVabsorbing capacities, thus hinting the similar penetration depth in UV region. Given the Dp is larger than 10 nm, the UV light will be partially absorbed by the 2─6 nm a-TiO2x(OH)y shell and then reach the core c-TiO2 in c-TiO2@a-TiO2-x(OH)y ( Figure R1 and R2). Therefore, both the core c-TiO2 and shell a-TiO2-x(OH)y will be excited by the UV light and then yield photo electrons and holes. We added the calculation results, schematic and corresponding discussion on page 10-11, line 273-290 and in Supplementary Fig. 11 and Fig. 12.  (2) Mulliken charge is coarse for calculating atomic charge, 0.01 e may be an error with this method. while the 0.002 Angstrom of O-H bond length difference is also maybe within the limit of error.
Reply: Agreed. We have deleted the coarse-accuracy (for screening purpose only as stated in the manuscript) Mulliken charge and geometric calculation in Fig. 3d to avoid any misunderstandings. Instead, we add a theoretic schematic to the SFLP model and a detailed analysis of the 1 H-NMR result to identify the SFLP protons, and focus on the high-accuracy Bader charge calculation (Supplementary Table 2) based on the amorphous structure to understand the charge distribution in the SFLP on page 13, line 337-340. "Theoretically, the terminal hydroxyl Ti(IV)-OH in the SFLP possesses higher degree of freedom compared to that of the bridging hydroxyl (Fig. 3d). Thus, their charge density and molecular vibrations should be distinct, which can be monitored through solid-state 1 H magic-angle spinning nuclear magnetic resonance ( 1 H MAS NMR) and attenuated total reflection Fourier transform infrared (ATR-FTIR), respectively. The 1 H MAS NMR spectrum of c-TiO2 showed peaks at 5.8, 1.6 and -1.65 ppm. The peak around 5.8 ppm is typically ascribed to absorbed water 1 , and the peak at 2.2 ppm on neat titanium dioxide is assigned to terminal hydroxyl group. Both water and hydroxyl group are naturally existing species when a metal oxide sample is exposed to air. The peak at -1.65 ppm appeared to be a background signal as similar signal emerged on the 1 H MAS NMR spectra of c-TiO2@a-TiO2-x(OH)y as well. Interestingly, the water and hydroxyl peaks of c-TiO2@a-TiO2-x(OH)y shifted to 5.2 ppm and 1.7 ppm, respectively, and two additional peaks (2.2 and 0.6 ppm) 2,3 emerged around the hydroxyl peak. Since the chemical shift of 1 H-NMR reflects the shielding effect, compared to c-TiO2, the lower chemical shift of water on c-TiO2@a-TiO2-x(OH)y indicated the possible electron donation from oxygen vacancies and/or Ti(III) in a-TiO2-x(OH)y, while the higher chemical shift of hydroxyl (1.7 and 2.2 ppm) suggested the proton of less electron density, which agreed well the Lewis base -OH in the SFLP model. The peak at 0.6 ppm can be assigned to several species, including terminal hydroxyl group and proton bonded to oxygen vacancy, which are closely related to the amorphous a-TiO2-x(OH)y shell (Fig. 3e)  Reply: Thanks for your comment, however, the experimental valence/conduction band positions and bandgap energies are already provided for both c-TiO2 and c-TiO2@a-TiO2-x(OH)y (Fig. 4), which are determined using a combination of UV-VIS-NIR, Tauc plot ( Fig. 4a and Supplementary Fig. 8) and UPS spectra ( Supplementary Fig. 9). Typically, theoretic band structure is calculated and plotted against high symmetry points of a crystal in DFT. However, in an amorphous solid devoid of symmetry, the physic meaning of band structure calculation is ambiguous, and thus it is not usually possible to determine a precise dispersion relation. Therefore, no theoretic band structure result is given but instead the TDOS (Fig. 4b, c and Supplementary Fig.10) to help understand the experimental results.
(4) From Fig. 4b,c, in the energy range of -1 -0 eV, there is a slight TDOS peak with high IPR for both a-TiO2-x(OH)y and TiO2-x, but not appear in the TDOS /PDOS results in Supplementary  Fig. 10.
Reply: We appreciate the referee's vital comment and apologize for the oversight in the Fig. 4b. We double-checked the result and found that the original Fig. 4b was the result for amorphous TiO2-x rather than the labeled amorphous TiO2-x(OH)y. Therefore, the TDOS in Fig. 4b was same to that in Supplementary Fig.10 (amorphous TiO2-x). We have revised Fig. 4b by a new calculation based on the amorphous TiO2-x(OH)y, and the new result ( Figure. R3) has been supplemented to replace the previous one. Now the TDOS in Fig. 4b agrees with that in Supplementary Fig.10a (TiO2-x(OH)y), and the major claim of "SFLP-facilitated charge localization" related to Fig. 4b remained valid. Fig. 4c represents the crystalline TiO2-x which does not present in the Supplementary  Fig.10, so there is no conflict. Figure R3. The total density of states (black line, right axis) and the corresponding values of the inverse participation ratio (IPR) (blue dots, left axis) for the a-TiO2-x(OH)y surface.
(5) Phonon calculations should be appended to provide the structural stabilities for both surface structures in DFT. Especially, the c-TiO2 structure in Fig.5 seems to be a monoatomic layer. Generally, this is unreasonable for surface structure.
Reply: We appreciate the referee's helpful comment for pointing out the possible misunderstanding in the DFT model. The c-TiO2 model used in this study is not a monoatomic layer but a slab structure with multiple atomic layers and a vacuum slab. The previous monoatomic layer-like  Supplementary Fig. 25. If a phonon frequency is positive, that means a positive curvature of the potential energy surface. So the energy increases quadratically if atoms are placed in the directions given by the associated eigenvector. Otherwise, the imaginary (or "negative") frequency indicates an energy decrease, or known as an "unstable" structure for a DFT model that does reach local free energy minimum. To this regard, phonon calculations are usually performed in transient-state search to verify the "stability" of intermediate models where conventional geometric optimization to search for the energy minimum is not applied.
However, in our steady-state, conventional geometric optimization, the structural stabilities of our crystalline and amorphous structures are promised by performing the standard DFT optimization via conjugate gradient minimization algorithm until the magnitude of residual Hellman-Feynman force on each atom was less than 10 -2 Ry/Bohr. This geometric optimization is superior to phonon calculation to this case. Furthermore, our amorphous TiO2 models and methodology to model amorphous TiO2 are also extensively validated in our previous works. 1 The authors presented the TOF of c-TiO2@a-TiO2-x(OH)y was high and superb among known catalysts, but did not provide the comparison. This manuscript presents the beneficial synergy between structural disorder and surface frustrated Lewis pairs in photocatalytic CO2 hydrogenation, with excellent CO production rate and TOF. It is helpful for designing SFLPsbase photocatalysts.
Reply: Thanks for the referee's comment, however, the comparison for CO2 photocatalysis is already shown in Supplementary Table 3. The activity of our catalyst, c-TiO2@a-TiO2-x(OH)y, is comparable to those achieved over supported metal photothermocatalysts in CO2-to-CO.
(7) From the Fig.6a, increasing the component of SFLPs will notably improve the CO production rate, so what is the role for c-TiO2 core?
Reply: Thanks for reviewer's vital comment on clarifying the mechanism. The SFLP is a concomitant in the Na-treatment of the c-TiO2 surface to generate the a-TiO2-x(OH)y shell. One feature of a-TiO2-x(OH)y is processing high concentration of oxygen vacancy. It is well known that oxygen vacancies can act as electron traps, and thus surface oxygen vacancies can facilitate surface reaction due to elongated lifetime of charge carriers, but dense bulk oxygen vacancies severely hamper the charge separation and transportation to reach the surface 1, 2 .
The efficiency of a photocatalytic reaction is determined by the efficiencies of light absorption, charge separation and finally the surface reaction. The surface reactivity and light absorption of a-TiO2-x(OH)y catalyst should be similar to that of c-TiO2@a-TiO2-x(OH)y when stacked together in the powder form, and the light penetration depth of the former should be smaller than the latter, as shown in Figure R5. Due to limited charge separation of a-TiO2-x(OH)y compared with the c-TiO2@a-TiO2-x(OH)y, the activity of a-TiO2-x(OH)y should be lower than that of the c-TiO2@a-TiO2-x(OH)y due to smaller surface area induced by the light penetration depth ( Figure R6). The whole picture of the mechanism is shown in Fig. 1 and Supplementary Fig. 12.   Fig. 1a and 1e (below picture, the orange arrows in Fig. 1b and 1e indicate the spatial electron transfer while the black arrows in Fig. 1a indicate the charge carrier transfer in terms of energy band levels). The effective mass is associated with the electronic band curvature along a specific direction, which is proportional to the second derivative of the energy band with respect to the wave vector k. Given the DFT calculation of effective mass of charge carriers are typically realized through mathematic calculation on top of band structure dispersion function, the amorphous structure devoid of high symmetry cannot get a band structure dispersion, nor to perform such a kind of calculation. Instead, the IPR analysis is provided to understand the carrier mobility (Fig. 4b).
Reference This manuscript created surface frustrated Lewis pairs (SFLPs) in an amorphous a-TiO2-x(OH)y surface. It discusses in detail the structure-activity relationship between the amorphous structure and the CO2 reduction reaction. This work is meaningful for studying the influence of surface disorder structure on photocatalytic CO2 reduction mechanism. It is a topic of interest to the researchers in the related areas, but the paper needs significant improvement before publication.
Reply: We appreciate the referee's positive comments on the novelty of our work. During this revision, we added additional experimental results (control activity tests), calculations (light penetration depth) and explanatory sentences to resolve concerns related to the mechanism. Actions taken are indicated in blue while explanatory sentences are shown in red. The peak around 5.8 ppm is typically ascribed to absorbed water, and the peak at 2.2 ppm on neat titanium dioxide is assigned to hydroxyl group. Both water and hydroxyl group are naturally occurring species when a metal oxide sample is exposed to air. The peak at -1.65 ppm appeared to be a background signal as similar signal emerged on the 1 H MAS NMR spectra of c-TiO2@a-TiO2-x(OH)y as well. Interestingly, the water and hydroxyl peaks of c-TiO2@a-TiO2-x(OH)y shifted to 5.2 ppm and 1.7 ppm, respectively, and two additional peaks (2.2 and 0.6 ppm) 1, 2, 3 emerged around the hydroxyl peak. Since the chemical shift of 1 H-NMR reflects the shielding effect, compared to c-TiO2, the lower chemical shift of water on c-TiO2@a-TiO2-x(OH)y indicated the possible electron donation from oxygen vacancies and/or Ti(III) in a-TiO2-x(OH)y, while the higher chemical shift of hydroxyl (1.7 and 2.2 ppm) suggested the proton of less electron density, which agreed well the Lewis base -OH in the SFLP model. The peak at 0.6 ppm can be assigned to several species, including terminal hydroxyl group and proton bonded to oxygen vacancy, which are closely related to the amorphous a-TiO2x(OH)y shell (Fig. 3e) Figure S7, weak peaks in the bridging region emerged on the c-TiO2 sample, while large peak intensities with increased peak numbers were observed on the c-TiO2@ a-TiO2-x(OH)y surface. Does the enhancement of the bridging OH feature mean the generation of Bronsted

acid [Ti(IV)-OH-Ti(IV)]? In this case, SFLP [Ti(III)-[O]-Ti(IV)-OH] and conventional Bronsted acid [Ti(IV)-OH-Ti(IV)] are co-existed on the c-TiO2 surface?
Reply: We appreciate the reviewer's comments and agree with the co-existence of bridging and terminal OH in c-TiO2@a-TiO2-x(OH)y. The generation of bridging OH is found to be oxygen vacancy-dependent H2O heterolysis, which is well-documented in literatures 1, 2 . The process can be summarized as Ti-[O]v-Ti-O + H2O = Ti-OH-Ti-OH. Given the [O]v-rich nature of a-TiO2-x(OH)y, the bridging OH generation is possible during the H2O washing or atmospheric moisture adsorption, besides the terminal OH generation. However, the bridging OH (pure c-TiO2) is inactive towards H2 splitting and less active for CO2 reduction compared to SFLP (c-TiO2@a-TiO2-x(OH)y). Reply: We appreciate the referee's vital comments on the mechanism. The light absorption of a material is determined by its extinction coefficient and light penetration depth. The penetration depth (Dp) of UV light for a-TiO2-x(OH)y is larger than 10 nm, so the UV light can penetrate the 3-6 nm shell to reach and excite the core TiO2. To amplify, the Dp = 1/, where  is the light absorption coefficient.  can be calculated from using the formula of  = 4πk/λ, where k is the extinction coefficient that is well documented for white TiO2, 1 λ is the wavelength of incident light. Based on above calculation, the wavelength-dependent Dp is listed below for white TiO2 ( Figure R1). According to the reflection spectra of black titania (Fig. 4a), its UV light absorption is close to the white TiO2, thus the similar penetration depth of several hundred nanometers in UV region. According to accessible absorption/refraction index data in black titania 2 , the visible light penetration depth of black titania is within several micrometers, so the visible light cannot be fully absorbed by a 3-6 nm surface a-TiO2-x(OH)y layer but will be absorbed by the subsurface catalyst particles with a stacking thickness reaching several micrometers ( Figure R2). We also added this part and the corresponding discussion on page 10, line 273-286 in Supplementary Fig. 12. Figure R2. Schematic of the light penetration depth and effective photochemistry surface area for c-TiO2@a-TiO2-x(OH)y and pure a-TiO2-x(OH)y in the stacked powder form. The total light absorption between the two samples should be the same which equals to (1-R), where R is the reflectance by the surface, but the Vis-IR light penetration depth of c-TiO2@a-TiO2-x(OH)y should be larger than that of a-TiO2-x(OH)y due to the Vis-IR transparent c-TiO2 core. Thus, the light accessible surface area of the former should be larger than the latter.
The role of amorphous shell in c-TiO2@a-TiO2-x(OH)y can be clarified by comparison with pure c-TiO2 or pure a-TiO2-x(OH)y. Pure TiO2 can only absorb UV light and is transparent to visible and infrared light, while pure TiO2-x(OH)y can absorb UV-Vis-IR light. In terms of converting the absorbed photons into charge carriers to drive chemical reaction, the pure TiO2 is superior in e --h + separation and subsequent transfer from bulk to surface compared to that in the TiO2-x(OH)y, since the bulk oxygen vacancies in the latter would act as the deep trap for electrons 3,4,5 .
The c-TiO2@a-TiO2-x(OH)y, on the other hand, can make the best use of the fullspectrum light-absorption capability of TiO2-x(OH)y and the charge-separation capability of TiO2 by forming a core-shell hybrid structure. This claim is further supported by a non-stoichiometry-dependent activity test in Reply to Q5, where the optimal x value of c-TiO2@a-TiO2-x(OH)y is 0.0015<x<0.0031 and further increasing the x value of the catalyst results in activity decrease.
Reply: We appreciate the referee's vital comment. The core c-TiO2 can only absorb UV light, while the absorbance of c-TiO2@a-TiO2-x(OH)y expands to visible and IR light (Figure 4a). 420 nm cut-off test helps to clarify the contribution of UV portion in the full Xe lamp spectrum, and demonstrates the visible and IR light utilization of the a-TiO2-x(OH)y shell. We have added corresponding discussion in page 15, line 413-420. 5. Figure 6a showed the more disordered structure, the stronger the photocatalytic performance. Is there an optimal value of reduction degree give the best photocatalytic performance?
Reply: We appreciate the referee's insightful comment on the control test. As shown in Figure R3b, the CO production firstly increased with non-stoichiometry of c-TiO2@a-  6. The CO rate under about 700 nm wavelength light should be measured and shown in Figure S18. Does the author consider the corresponding apparent quantum efficiency(AQY)?
Reply: We appreciate the reviewer's comments and apologize that our LED sources don't cover > 700 nm wavelength. Instead, we supplemented the spectra of our red, green and blue LED ( Figure R4) and calculated the AQY. Where the light intensity is 4 W, illumination area is 1 cm 2 , I% is the percentage of the Xe light intensity at certain wavelength ( Figure R5), A% is the light harvesting efficiency at certain wavelength according to the absorption spectra ( Fig.4a  Reply: Appreciate the reviewer's comments. Long-term stability (48h) was shown in Supplementary Fig.16 and CO production kept steady. Theoretically, the surface OH can be recycled through H2 splitting and oxygen vacancy can maintain via H2 reduction. O1s XPS spectra for the catalyst before and after reaction evidenced the population of both [O]v and hydroxyl group only decreased slightly ( Supplementary Fig.18), which is in good agreement of the stability test in Supplementary Fig. 16. 8. The band structure can be changed by the introduction of an amorphous shell. However, the bandgap of the materials is not suitable for photocatalytic. Considering the previously reported excellent band gap of black titanium oxide, the advantage of this structure should be discussed.
Reply: Agreed. The optical properties of semiconductor nanomaterials strongly depend on their band gaps. TiO2 can be used for photocatalytic CO2 reduction because of its suitable valence band (VB) and conduction band (CB) levels. 1, 2 Narrowing the bandgap of the TiO2 will reduce its redox capability to exchange for the enhancement in light absorption. To this regard, previous black titanium oxide with reduced band gap 3 should be excellent only for water splitting, instead of the CO2 catalysis that requires high reduction potential. The advantage of our crystalline core-amorphous shell is 1) increasing the light transportation path, 2) and thus increasing the effective surface area for photochemistry, 3) inheriting high reduction potential from TiO2, 4) combining the advantages of both TiO2 and a-TiO2-x(OH)y in charge separation and light absorption, respectively. These merits promote the catalytic efficiency towards photo(thermal)catalytic CO2 reduction, as evidenced by the activity test. Details can be found in Reply to Q3.

The influence of the content of the amorphous structure or the [Ti(III)-[O]-Ti(IV)-OH]
bonds should be explained. If there is optimal content, the paper should present the results.
Reply: Appreciate the reviewer's comments. Based on the formation of Ti(III)-[O]-Ti(IV)-OH via the TiO2-x+ 0.5yH2O  TiO2-x(OH)y reaction, the amount of SFLPs corresponds to the measured x. Similar to the Reply 5, we have discussed the optical stoichiometry was about 0.0015<x<0.0031 for photocatalytic CO2 hydrogenation. Further increasing the non-stoichiometry with the thicker amorphous shell ( Figure R7) will lead to electron trapping in the bulk and lowered activity in CO2 hydrogenation ( Figure R3b). Although increasing the amount of SFLPs may create more active sites, more thicker shells will limit the charge transportation. To make full of their synergy between surface chemistry and photocatalysis, the sample c-TiO2@a-TiO2-x(OH)y (0.0015<x<0.0031) exhibited the best performance in this work. We also added the corresponding description on page 15, line 391-406 in manuscript and the figure in Supplementary Fig.  22. Figure R7. HR-TEM micrograph of c-TiO2@a-TiO2-x(OH)y with increasing stoichiometry. The amorphous/crystalline interfaces are marked with dotted lines.
The revised manuscript has been well addressed based on the proposed comments. It could be accepted in its present form.
Reviewer #2 (Remarks to the Author): This work investigated the interrelationship between structural disorder and surface frustrated Lewis pairs (SFLPs) in a core-shell c-TiO2@a-TiO2-x(OH)y heterostructure for CO2 hydrogenation photocatalysis. The results of this study are helpful for constructing efficient photocatalytic reduction of CO2 by titanium oxide, which can provide researchers with a more in-depth understanding of the mechanism of photocatalysis. As the authors claimed in the paper, the amorphous layer plays a very significant role in this photocatalysis process. The authors demonstrated that the surface of the titanium oxide contains amorphous layers by HRTEM technology. However, the HRTEM is just a local characterization method. At the same time, the surface of titanium oxide is easily damaged by electrons and forms amorphous layers. Based on these considerations, I suggest that the authors need use other powerful methods to jointly show that the surface of titanium oxide contains amorphous layers.