General acceptance of perovskites for organic synthesis

Perovskite colloidal suspension (CsPbX3: ECB = −1.2~−1.4 V, EVB = +0.6~+1.5 V, all vs SCE; CB: conduction band; VB: valence band)23 are effective catalysts for several fundamental organic reactions under visible light as shown in Fig. 1. Direct C–C bond formations are observed via C–H activation of aldehydes (1a, 1b) or tertiary amines (1c, 1d). The scope of the former reaction is not only limited on previously explored C–Br weaker bonds17, but also covers stronger C–Cl bond. The absence or presence of oxygen is the key to lead to chain-extension product (1c) or an unexpected cyclization reaction (1d). C–N bond formations via direct N-heterocyclizations forming pyrazoles (2af) and pyrroles (2gi), critical reaction for pharmaceutical development, are realized in high yield with perovskite at room temperature. C–O bond formation via aryl-esterification (3af) was achieved with a Ni co-catalyst. The respective reaction conditions are also optimized with regards to solvents, types of perovskites, air-tolerance, co-catalysts, and reaction time, etc. (see Supplementary Tables 18 for details). Catalyst loading has also been explored (Supplementary Tables 17) and respective minimum loading for typical reactions of ~0.1–0.5 mmol has been listed in Fig. 1. These reactions result in respective products in moderate to high yields without need for anaerobic sparging. The scopes of each aforementioned reaction were explored with various functional groups. (Fig. 1 and “Methods” section for details) As expected, control experiments reveal no product in the absence of photocatalyst or light.

Fig. 1


The library of C–C, C–N, and C–O bond formation reactions and respective yield. (Yields of 1a, 1c, 1d, 2a, 2g, 3a are the average yields of three reactions, see Supplementary Table 8; Inset: perspective view of 1d’s single crystal structure with the thermal ellipsoids drawn at 50% probability level and the H atoms omitted for clarity.)

Perovskite’s size effect

The perovskite colloids, P1, described above are readily synthesized according to previous report17,24 via directly mixing of readily available low-cost starting materials, PbX2 with CsX, in an open vial under bench-top conditions (Supplementary Fig. 1). The resulting gram-scale emissive perovskite colloids exhibit a broad size-distribution, ca. 2~100 nm (Fig. 2a). The observation together indicates a bandgap energy of 2.4 eV that well matches the bulk CsPbBr3 bandgap7,25. The synthesized colloids are too large to be in the quantum-confinement regime (Fig. 2a). Thus, for the system we are considering most colloids within the ensemble are larger than the Bohr radius, and hence the bandedges are determined by bulk bandedges and quantum-confinement effects do not contribute.

Fig. 2


Characterization and spectroscopy studies of photocatalysts. a TEM of CsPbBr3P1; b P3; c P4; d UV-vis and PL spectra of CsPbBr3P2P5; e PL spectra for P1 and P4 in CH2Cl2 as prepared and after LED irradiation for 24 h and 1 h, respectively. f XRD of as-prepared CsPbBr3P1; isolated from the reaction 1a before and after irradiation, respectively; g the corresponding XRD for reaction 1b; h PL spectra of P1 in THF with addition of TFA; i PL spectra of CsPbBr3 NCs, Ir(ppy)3, CdSe QDs and Ru(bpy)3Cl2 in air or N2-saturated solutions. Source data are provided as a Source Data file

In contrast, using a high temperature synthetic method26,27, we also synthesized size-controlled CsPbBr3 NCs (P2 14 nm, λPL = 521 nm; P3, 9 nm, λPL = 515 nm; P4, 6 nm, λPL = 508 nm; P5, 4 nm, λPL = 467 nm, Fig. 2b–d and Supplementary Fig. 2). As shown in Fig. 2d, these NCs show a blue-shift probably due to quantum confinements26,27. The photocatalytic ability has also been explored in the same reaction condition. In C–H activation, at the early stage of the reaction, we find that smaller size NCs, i.e. P2P4 show a higher initial reaction rate compared to the original synthesized P1 NCs. (Supplementary Fig. 3). However, small size NCs’ catalytic reactivity diminished quickly. When breaking a C–Br bond to form 1a, the reaction yield is recorded as 54–64% using P2P4 in less than 40 min, and longer reaction time leads to a marginal increase of the yield of 1a. Much lower yield, ~8% was observed within P5 probably due to a significant blue-shift leading to less visible absorption. Whereas using P1, the reaction rate is slower, however, the yield continuously increases and reaches 85% in ca. 5 h.

We suspect that small size NCs have higher surface area-to-volume ratio (Supplementary Table 9), hence a faster rate at the early stage. However, detrimental effects, i.e. moisture residue in solvent are inevitable. Such effects are more prominent on small size NCs than P1. We assume if the desired photocatalysis is slower than perovskite decomposition, the reaction yield may be of significant discrepancy between small and large size NCs. Such assumption is corroborated with reaction 1a described above. In contrast, if the decomposition is not prominent, the yield discrepancy is less obvious. In fact, in 2a, perovskite is stable in a pre-dried non-halide solvent ethyl acetate (Supplementary Fig. 4). 2a is produced in 86% yield with P2 in 2 h, 87% using P1 in 6 h (Supplementary Fig. 3). Overall, small size NCs, in general, promote a faster reaction rate, but not necessarily a higher yield unless presenting in a perovskite friendly reaction environment. Considering synthesis merits, large size NCs, in general, provide higher yield although a longer reaction time in a scale of 6 h or higher is required.

Stability and reaction condition tolerance

Pb-halide perovskites’ photovoltaic performance perishes over moisture28,29, impeding the wide commercial application of such materials as solar cells. The stability is quite distinct if perovskites are to be applied to organic synthesis in which more critical parameters may influence the stability of perovskites, i.e. solvent type, ions, acidity, etc., and further manipulate the catalytic ability. Thus, these parameters are evaluated individually for a better understanding of perovskite photocatalysis. A quite strong stability of P1 in organic solvents was indicated by no obvious PL changes of CsPbBr3 for several weeks in less polar organic solvents13,17. (Note that P1 is not stable in polar solvents, i.e. acetone, acetonitrile, DMF, DMSO, Supplementary Fig. 4). However, P2P5 are less prominent and significant PL diminishing is observed. (Supplementary Fig. 2) Interestingly, under the irradiation of LED, PL blue-shift of P1 in CH2Cl2 are observed in 24 h. (Fig. 2e) Such changes are significantly magnified on P2P5 as shown in Fig. 2e and Supplementary Fig. 5, absorption and PL blue-shift within in 1 h, whereas no obvious PL changes are observed in non-halide solvents. This observation may be attributed to a photoinduced fast halide exchange for CsPbBr3 with CH2Cl2 as previously reported16,30.

Next, we evaluate the ion effect in perovskites’ photocatalysis. Perovskite is reported to sensitive to both inorganic cations and anions31,32,33,34. In our photocatalytic setup, co-catalyst (ClCH2CH2Cl)2NH2Cl in reaction 1a, leads to an initial PL blue-shifted due to anion-exchange forming CsPbBrxCl3-x, confirmed by XRD (Fig. 2f). It is interesting to point out that co-formation of Br ion during reaction 1a, may further exchange with the CsPbBrxCl3-x and stabilize the perovskite NCs. Such stabilization is evidenced by the after-reaction catalyst characterization in which XRD indicates that the isolated photocatalyst solid was corresponding to CsPbBr3 and surprisingly, no peak has been assigned to CsPbCl3 (Fig. 2f). This is probably because the co-formation Br ions are in chemical equivalency and its concentration is significantly higher than that of Cl. Therefore, a Br compensated and stabilized CsPbBr3P1 photocatalyst system is thus observed. (Supplementary Fig. 6) In contrast, reaction 1b employing Cl-substrates leads to a fully-exchanged CsPbCl3 after reaction (Fig. 2g). Overall, perovskite P1 shows a much better stability during the reaction 1a, in which the NCs can be isolated from the previous reaction mixture via centrifuging and then re-suspended for a new reaction under identical conditions for at least four cycles with slightly PL blue-shift, whereas small NCs P4’s recycling ability is limited (Supplementary Fig. 3). As comparison, when free halide anions are absent, for example in reaction 2a in EtOAc solution, the overall stabilities for P1 and P4 are enhanced and result in an improved recyclability in such perovskite friendly environment (Supplementary Figs. 6 and 7).

Acidity or free protons in perovskite reaction mixture may play a role in organic synthesis. For instance, carboxylic acids such as propionic acid, benzoic acid or trifluoroacetic acid (TFA), were employed as the co-catalyst (1c and 1d) or as a substrate (3a3f). Thus, we first measured the PL for perovskite NCs with different acids to elucidate the tolerance of acidic conditions. Interestingly, as shown in Fig. 2h and Supplementary Figs. 8 and 9, a PL enhancement of P1 was observed upon the addition of benzoic acid, propionic acid, and also TFA (see Supplementary Movie 1). This is corroborated with previously observed PL enhancement using thiophenol16, phosphoric acid35 etc. The PL enhancements are probably because carboxylic acid function as the capping ligand by the strong hydrogen bonding with surface halide ions35 and may also account from a strong interaction between carboxylic acid and Pb atoms, indicated by Tan et al.36. Acid binding with defects on perovskite may also lead to an enhanced PL performance according to Zhu et al.37. The maximum PL was observed using TFA at a concentration of ca. 6.5–13 mM, more acid leads to a diminishing PL probably because large number of protons may start to initiate a deactivation process. Interestingly, such optimized TFA concentration also leads to a maximum product yield of 1c and 1d as shown in Supplementary Tables 3 and 4, indicating that a high PL of the photocatalyst may increase the catalytic conversion. Therefore, non-halide organic acid may not only stabilize the perovskite NCs, but also may increase the overall catalytic efficiency for respective reactions.

Key catalytic parameter comparison with other photocatalyst

Air-tolerance is important for the practical end-use of chemical synthesis. One distinct advantage of our colloidal system is that the organic reactions observed here occur without the need for N2-sparging. In stark contrast, molecular photocatalyst38 necessitates air-free reaction conditions. The key difference here is that the perovskite NCs likely undergo faster quenching from the organic substrates, while quenching from air is negligible. (Fig. 2i and Supplementary Fig. 10) The reverse is true for most cases of molecular catalysts – quenching is substantial an., O2 quenching is substantial and competitive with the catalytic reactions, leading to poor catalytic results. Hence, yields of reaction 16,39,40,41, 242,43, and 344 in air with perovskite are significantly higher than with others. (Table 1, Supplementary Tables 17) For instance, 1a were obtained in 85% yield in air using perovskite, but only resulted in trace amount with Ru(bpy)32+. These results suggested that perovskite may exhibit a broad tolerance, particularly towards air.

Table 1 Comparison of photocatalysts for corresponding reactions in air or in oxygen

Catalytic turnover number (TON) is compared and listed in Table 1. Heterogeneous catalyst, i.e. 3.0 nm CdSe QDs were reported to optimally render a TON of 79,100 (based on QD’s molecular weight Mw, 88,000 g mol−1) in glove box45. However, in our condition under air, no yield (nor TON) of 1, 2, and 3 can be obtained using CdSe QDs. In addition to air-sensitivity, CdSe’s performance was also dependent on size and capping ligands45. While changing capping ligand on perovskite plays little role in the yield as shown in Supplementary Tables 24. This is probably because the capping ligands (e.g., n-octylammonium) that stabilize perovskite colloids are reported to function as A site to the perovskite APbX3 structure31, hence no extra stabilization protocol is required using perovskite nanocrystal for photocatalysis. Using the method in CdSe QDs45 to calculate TON, P2 NCs (14 nm, based on Mw, 8,015,000 g mol−1, P1P5 TON see Supplementary Table 9, calculation details see Supplementary Note 1) renders 2,565,000. Perovskites’ heterogeneous catalytic ability is validated via regaining strong PL after recovering the catalyst via centrifuge after reaction (Supplementary Fig. 7). To compare TON with molecular catalysts, TON calculation based on mole of metal (independent of size, CsPbBr3, 579.8 g mol−1) was carried out instead. For instance, four cycles of the reactions afford a TON, at least 9,100 for 1a (Table 1, details see Supplementary Note 2). Overall, one or two orders of higher TONs under our condition are observed using perovskite than others, except reaction 3 in which TON may rely on both perovskite and Ni co-catalyst.

Higher activity of perovskite than other photocatalysts may account from the intrinsic photophysical properties on charge separation and transfer. For example, the perovskite NC’s ultrafast interfacial electron and hole transfer dynamics has been revealed by Lian et al.46. First, negligible electron or hole trapping has been found in perovskite NCs, facilitating photoredox catalytic cycle. In the presence of organic substrates (as electron or hole acceptors in photoredox organic synthesis2), photon-induced excitons in perovskite can be efficiently dissociated and separated46. For instance the half-lives of electron transfer to an organic electron-acceptor is reported to be ~65 ps, while charge recombination rate is reported about ~2 orders slower. The hole transfer dynamics from perovskite to an organic substrate is also reported to be 20 times faster than its recombination46. Such observation is also corroborating with our previous reports on the ultra-slow recombination velocity of perovskite both in CsPbBr3 and CSPbI3 single crystals and films8,18,19,20. Overall, the lack of electron and hole traps and fast interfacial electron transfer and hole transfer rates are imperative that may enable highly efficient perovskite induced photocatalysis. In fact, the superior performance is not surprising given that when employed in photovoltaics, the Pb-halide perovskites also perform much better (PCE, 24.2%) compared to transition metal-based dye-sensitized solar cells (11%), QD photovoltaics (12%) and organic photovoltaics (12%)12.


Oxygen may be of an essential component in certain photoredox reactions. For instance, in Fig. 3a, radical addition product 1c is achieved in nitrogen atmosphere while in a similar setup, air or oxygen atmosphere produces a ring-closure 1d (crystal structure provided in Fig. 1). Oxygen is found to be the key reagent as the hydrogen atom acceptor that further induced the C–H activation on phenyl rings39,40. As shown in Fig. 3, the reaction mechanisms are proposed in which the key radical intermediates have been investigated. Upon Stern–Volmer studies (Supplementary Figs. 1117), perovskite PL quenching by 1d-A was observed (kq = 3.6 × 108 M−1 s−1, details see Supplementary Fig. 12 and Supplementary Note 3) and resulted in 1dB radical in the presence of oxygen. Intermediate 1d-B and 1d-C have been verified via radical trapping experiment employing 2,2,6,6-tetramethyl-1-piperidinyloxy (TEMPO) as a radical scavenger, through LC-MS (Supplementary Figs. 19 and 20). In the absence of oxygen, radical 1c-B is also confirmed by TEMPO-trapped product (Supplementary Fig. 18) and further verified by the self-coupling 1c-C via 1H NMR. It is worth mentioning that the presence of air leads to more 1c-C formation and ultimately diminishes the yield of 1c.

Fig. 3


Mechanisms. a Proposed mechanisms for the synthesis of 1c and 1d; b 2a and 2g. (Blue square: isolated and characterized by 1H-NMR; Red square: trapped and detected by LC-MS (Supplementary Figs. 1823); HT = hole-transfer; ET = electron-transfer)

Figure 3b shows the proposed mechanism of C–N formations, in which both oxidative (ET, 2a-A) and reductive quenching product (HT, 2a-B) in reaction 2a have been trapped by TEMPO (either observed via 1H NMR or LCMS), indicating a strong charge separation and transfer ability induced by perovskite. This pathway is similar to our previous mechanism exploration in α-alkylation of aldehydes17. Radical coupling between 2a-A and 2a-C leads to the intermediate of 2a-D. Then C–N formation via intramolecular cyclization and a final dehydration leads to the pyrazole product 2a. In contrast, the radical formation from 2g-B via direct HT has not been observed, instead 2g-C was verified via radical-trapping, likely demonstrating a different mechanism of pyrrole formation as shown in Fig. 3b.

To further elucidate the reaction mechanism, electrochemical studies were conducted. (Supplementary Figs. 2431) According to the comparison between redox potentials of the key substrates and the band energy of perovskite, the respective driving force is listed in Fig. 4. Driving force for HT in reaction 1c, 1d and 2a is observed among ~0.1 to 0.3 eV, consistent with the Stern–Volmer quenching results (Supplementary Figs. 1117) as well as the mechanistically verified intermediates in Fig. 3. However, 2g-B disfavors HT due to a more positive oxidation potential (Eox, 1.42 V vs SCE), corroborating with the previous observation that direct radical forming from 2g-B is difficult, unlike reaction 2a pathway. Moreover, driving force for ET is also listed from ~0.2 to 0.5 eV, confirming our discussion on ET in Fig. 3. However, noticeable exception, 2,4′-dichloroacetophenone, though presenting a more negative reduction potential (Ered, −1.47 V vs SCE), still reacts to form respective pyrrole. We postulate that in-situ band-tuning of perovskite may play a role here and is discussed below.

Fig. 4


Band energy of CsPbBr3 vs the redox potentials of substrates. Source data are provided as a Source Data file

Unique band-tuning of perovskite

As discussed above, the perovskite NCs P1 are too large to be in the quantum-confinement regime and the majority of the NCs within the ensemble are larger than the Bohr radius. Thus, the band energy of our photocatalyst, analogs to excited state redox potentials, E* in molecular catalyst, is determined by the bulk bandedges. Bandedge-tuning is achievable by simply mixing of different ratio of halides32,33,47. We also observed that in-situ ion exchange using P1 results in band-tuning (Fig. 5a). In theory, as shown in Fig. 5b, c the bandedges of perovskite after tuning covers most of the E of the known Ru or Ir molecular photocatalysts.

Fig. 5


Band-tuning of perovskite. a The PL spectra of colloidal CsPbBr3 in dichloromethane via band tuning with trimethylsilyl chloride or iodide and their representative images under UV lamp (top). b Bandedges of APbClxBryI3-xy. c Excited-state potential (E*) range of APbClxBryI3-xy comparing with noble transition-metal catalysts. d Two successful reaction examples with perovskite band-tuning. Source data are provided as a Source Data file

The band-tuning is of critical importance for a photocatalyst to activate different types of substrates. For example, C–O formation reaction 3 is also proposed and shown in Supplementary Fig. 32 similar to previously reported mechanism44. It is reported that energy transfer from triplet excited state of Ir photocatalyst is the key for Ni complex activation thus resulting in an efficient reductive elimination for C–O bond formation44. Triplet energy (ET) exploration from Ir(ppy)3 derivatives via modifying the substitution group on ppy ligand demonstrated that a higher correlation between ET and the production yield. Specifically, a higher ET results in a higher yield. As shown in Fig. 5d, in our perovskite system, 3f is produced in trace amount if CsPbBr3 is employed with dtbbpyNiBr2 co-catalyst, comparing to 78% with dtbbpyNiCl2. While in Ir photocatalysis, the different halides on Ni co-catalyst only play a marginal effect44. We suspect that an in-situ ion-exchange from NiCl2 may result in a blue shift of perovskite, similar to the increasing ET in Ir system, thus leading to a significantly higher yield of 3f using co-catalyst dtbbpyNiCl2. To further confirm such hypothesis, we have conducted a systematic band-tuning experiment to demonstrate the correlation between the bandedges and the yield of 3f. In a typical experiment, perovskite CsPbBr3 is employed with NiBr2 co-catalyst, but tuned using a reported agent, i.e. trimethylsilyl chloride (TMSCl)34. We find that shifting the bandgap to higher values, by mixing with chloride to form CsPbClxBr3-x, increases the yield of 3f, similar to elevate ET in Ir system. However, more Cl component is not always beneficial for this type of reaction. As shown in Fig. 5a, PL intensity is significantly lower when Cl is incorporated into perovskite. Higher bandgaps (shorter PL peak wavelengths) resulted in a lower yield, and is likely tied to the lower PL quantum efficiency that indicates a competitive carrier trapping mechanism32. Overall, a maximum yield of 85% was obtained when the PL peak corresponds to 498 nm (Supplementary Table 7). This observation illustrates that the intentional band-tuning of perovskite NCs may activate previously non-reactive substrates.

Furthermore, band-tuning may also result in an absolute discrepancy in photo-activation. It is widely accepted that the C–Cl bond are stronger than C–Br and hence harder to activate48. Surprisingly, in reaction 2j in Fig. 5d, α-chloroketone is observed to react to form 2j in dioxane in a yield of 67% while α-bromoketone is almost non-reactive at all. We assume that the band energy of CsPbBr3 is not adequate to activate either Cl or Br-substrates in dioxane. However, ion-exchange may not only occur between CsPbBr3 and CH2Cl2 as previously reported16,30, but may also between CsPbBr3 and suitable organic Cl-substrates. Interestingly, CsPbCl3 was confirmed by XRD after reaction. (Supplementary Fig. 33) Cl-substrate is the only Cl source and hence is activated in this type of reaction. Hence CB of photocatalyst is thus moved higher, simultaneously the reduction potential of the substrate moves in a reverse direction (Fig. 5b), overall making the ET possible and finally resulting in the pyrrole formation. This is also corroborating with the observation that, if reaction 2j was conducted in CH2Cl2, CsPbBr3 NCs may first initiate ion-exchange with the solvent hence shift band energy, and finally catalyze both Br and Cl-substrates towards 2j formation. This result demonstrates that in-situ band-tuning of perovskite NCs may provide unexpected activity towards previously unachievable substrates.