Introduction

Extracting hydrogen (H2) from pure water powered by clean and inexhaustible sunlight is a long-dreamed solar energy conversion strategy in a way that compensates for the intermittency of solar irradiation1,2. Such technology requires a photocatalyst of superior visible-to-NIR light response that is maximally close to the solar spectrum, long-enough exciton lifetimes for ns-scale water splitting, and proper band structure for both hydrogen and oxygen evolution reactions (i.e., HER and OER)3,4. Although much desired, current-developed semiconductor catalysts, including transition-metal oxides5, -oxynitride6, metal-organic hybrid framework-based materials7, and metal-free inorganic semiconductors (e. g., C3N4 and BN)8,9, cannot meet all the requirements for commercial application with >10% solar-to-hydrogen (STH) conversion efficiency10,11. Besides the commonly low apparent quantum yields (AQYs) at the wavelength ≥420 nm visible-to-NIR region of incident sunlight4, current stereotypical photocatalyst designs require the loading of expensive or poisonous noble-metal/metal-oxide as co-catalysts to help release H2 and O21,2,3,4. In this sense, metal-free all-organic catalysts capable of solar overall water-splitting hold research perspectives11.

Compared with metal-based inorganic or organic-inorganic complex photocatalysts, covalent C-H-X organic materials available for various structural tailor and functional designs show superior advantages in light-harvesting initialed by π → π or n → π transitions11. Some even reach the world record for HER half-reaction efficiency over inorganic catalysts12. However, due to the fast radiative recombination of direct S0 → S1 excitations13,14 and the thermodynamical energy requirement of overall water-splitting4, direct STH conversion over common organic materials is impossible with low-frequency photons around 420–1000 nm, ~50% energy of the incident solar spectrum15. As a solution, the upconversion strategy based on the triplet-fusion can double the incident photon energy, driving the water-splitting reaction that is thermodynamically impossible to direct low-frequency irradiation16. Compared to the Kasha-allowed singlet excitation (S0 → S1), the triplet-fusion pathway is mediated by a long-lived triplet-triplet (TT) exciton pair16, which is inaccessible by a direct S0 → T1 transition due to the disallowed spin13. Consequently, dark-state TT pairs of an organic chromophore in a sensitizer-annihilator system can only be indirectly populated either by the S1 → Tn intersystem crossing (ISC, n ≥ 1) (Fig. 1a) or the S0S1 → 1(TT) → (TT) singlet fission (SF) process14,17,18, which generally requires strong UV/vis absorption to produce the initial singlet excited state on the sensitizer and a robust conjugation of the sensitizer-annihilator chromophore pair19. This is incompatible with utilizing full-band low-frequency photons in the solar spectrum. Moreover, as a competing pathway to the Dexter triplet-triplet energy transfer (TTET), high-energy triplet excitons arose from little S1/T1 energy difference are easily quenched by triplet O220, the main product of the water-splitting reaction (H2O → H2 + O2), to produce singlet O2, rendering STH conversions unsustainable. In addition, due to the common high overpotential on simple hydrocarbon surfaces, the H+ → H2 evolution reaction (HER), a key half-reaction in water splitting, can hardly occur on organic catalysts unless loaded with noble-/transition-metal-based cocatalysts.

Fig. 1: Schematic comparison between different mechanisms.
Fig. 1: Schematic comparison between different mechanisms.
Full size image

a The classic triplet-fusion upconversion pathway in an organic sensitizer-annihilator system through ISC, and b in this work, the two-band excitation upconversion pathway in a spin-hybrid :PDI2-/PDI2- all-organic semiconductor to utilize the full vis-to-NIR solar spectrum.

In contrast, a triplet-fusion pathway with TT pairs from intrinsic S0 → T1 transition can harness a wide range of incident sunlight, including near-infrared (NIR) photon energy, and be immune to O2 quenching with low triplet energy levels. Until now, such a spin-forbidden S0 → Tn transition, however, has only been achieved in a few examples by spin-orbit coupling with heavy atoms (e.g., Lanthanide, Iodine, Osmium, Platinum, and Gold)21,22,23,24,25. Although much desired, all-organic material-induced direct S0 → Tn transition without any heavy atoms has yet to be identified. We envision the direct S0 → T1 transition by a robust spin-orbit coupling between the preset open-shell triplet chromophore (i.e., diradical) and the singlet closed-shell isomer chromophore within a heavily conjugated crystal. Such a design not only makes the full use of the visible-to-NIR solar spectrum by the two-band excitation but also creates a proton-favorable electron transfer channel by diradical-dominated reactive interface on all-organic catalysts to break through the HER bottleneck (see Fig. 1b).

Results

Synthesis and catalyst structure

We opt N, N’-di(propanoic acid)-perylene-3, 4, 9, 10-tetracarboxylic diimide (denoted PDI throughout the article) (Fig. 2g) as the model perylene diimide parent to prepare the target spin-hybrid all-organic semiconductor (for details of the preparation methods, see Method). The pristine PDI in aqueous solution exhibits broad visible-light absorption with double peaks at 460 nm and 596 nm (Fig. 2a). After the hydrazine reduction (N2H4 + 2PDI → N2 + 2 :PDI2−/PDI2− + 4H+), :PDI2− diradical anions and corresponding PDI2− quinoid tautomer are generated with an emerging sharp absorption profile peaking at 530 nm and 575 nm. Distinct color difference between PDI and :PDI2−/PDI2− aqueous solutions can be distinguished (Supplementary Fig. 2). Electron spin-resonance (ESR) spectra show an intense radical spin feature at g = 2.0032 (Fig. 2b), indicating a triplet diradical feature of :PDI2−. A number of qualitative spectral characteristics (Supplementary Figs. 2–7) verify the :PDI2−/PDI2− formation and the structure with reduced cis-/trans-carbonyls (Fig. 2g) and the simplest protonation form of (H+)2-(:PDI2−/PDI2−). Note that after the two-electron reduction, singlet photoluminescence (PL) emissions at 540 nm and 585 nm corresponding to the radiative relaxation of S1 → S0 can be observed over the :PDI2−/PDI2− solid sample (Fig. 2d), suggesting the coexistence of closed-shell PDI2− quinoid isomers (S0) and open-shell :PDI2− diradicals (T1) through tautomerism. Compared to PDI, :PDI2−/PDI2− exhibits a much smaller Stokes shift (~15 nm) (Supplementary Fig. 8), suggesting a smaller reorganization energy than that of PDI. This is further demonstrated by smaller changes and shifts in infrared vibrational intensity on :PDI2−/PDI2− than PDI under the same excitation condition (Supplementary Fig. 9). Triplet states predominantly decay through non-radiative processes or much slower radiative processes like phosphorescence, resulting in the non-fluorescent :PDI2− diradicals as a ‘dark state’. Upon reduction by N2H4·H2O, the fluorescence peak of PDI rapidly decreases to zero within 10 s, resulting in the formation of non-fluorescent :PDI2− triplet diradical anions (Fig. 2e). However, within the next 30 min, fluorescence peaks at 646 nm and 690 nm gradually emerge, indicating the transition from non-fluorescent :PDI2− to fluorescent PDI2− via tautomerism transformation. Such a tautomeric conversion of ːPDI2− → PDI2− after the 2e− reduction of PDI was further in-situ tracked by the increasing PL emission in 30 min (Fig. 2e). After completely removing residual hydrazine hydrate from the solvent, singlet emissions of the quinoid recovered to 540 nm and 585 nm (Fig. 2f), identical to the solid sample (Fig. 2d). This means it is the co-existed hydrazine hydrate that makes the peaks of PDI2− significant redshift from 540 nm and 585 nm to the 641 nm and 688 nm, as shown in Fig. 2e. To further validate this, we performed additional control experiments by adding varying amounts (0~0.35 μL) of hydrazine hydrate to the DMF solution of the DMF solution of isolated :PDI2−/PDI2− sample. As expected, we observed a gradual redshift in the PL emission of :PDI2−/PDI2− (Supplementary Fig. 10), validating that the presence of hydrazine hydrate in the solution indeed causes a significant redshift in the fluorescence of :PDI2−/PDI2−. The diradical/quinoid ratio is determined by the integral PL emission intensity of :PDI2−/PDI2− compared with PDI (:PDI2− is non-fluorescent), suggesting a :PDI2−/PDI2− ratio of 74% to 26% in the precursor samples at room temperature (15 °C) (Fig. 2f). The solid :PDI2−/PDI2− precursor for subsequent self-assembly can be obtained by evaporating the solvent and stabilized under anaerobic conditions.

Fig. 2: Characterizations of :PDI2−/PDI2− precursor.
Fig. 2: Characterizations of :PDI2−/PDI2− precursor.
Full size image

a UV-vis absorption spectra of PDI and :PDI2−/PDI2− precursor in aqueous solutions. Inset shows the color comparison of different solutions. b ESR spectra of PDI and :PDI2−/PDI2− precursor. c Photographs of PDI and self-assembled :PDI2−/PDI2− nanobelt (120 h). Inset shows the magnified optical microscope image of the self-assembled :PDI2−/PDI2− nanobelt (120 h). d The PL spectrum (λex = 450 nm) of :PDI2−/PDI2− solid precursor sample. e In-situ PL spectra (λex = 450 nm) tracking the emission of PDI after the 2e− reduction by hydrazine hydrate in DMF solution in 30 min. With the addition of 0.25 μL hydrazine hydrate (85%), PDI was first converted to non-fluorescent :PDI2− diradicals. A tautomeric equilibrium from the non-fluorescent diradicals to fluorescent quinoids was then achieved in 30 min. f Comparing PL spectra (λex = 450 nm) between pristine PDI and :PDI2−/PDI2− samples from in-situ 2e− reduction of PDI for 30 min before and after removing residue hydrazine hydrate in DMF solutions. The hydrazine hydrate component was removed by evaporating the solvent by N2 flow and then redissolving the sample in an identical concentration. Without hydrazine hydrate, :PDI2−/PDI2− sample in DMF shows identical PL emissions to the :PDI2−/PDI2− solid sample as in d. g Schemes show the two-electron reduction pathway of PDI molecules generating :PDI2− diradical anions (T1) and the tautomeric PDI2− quinoid isomer (S0).

Due to the solid π-π interaction of perylene rings between individual PDI molecules, the simple phase-transition method has enabled the controllable self-assembly of single-phase PDI-based organic semiconductors with regular morphologies and delicate crystal structures (Supplementary Fig. 11)26,27. Applying the same self-assembly strategy, the self-assembly of :PDI2−/PDI2− was performed through the phase conversion in an optimized mixture of ethanol (poor solvent) : water (good solvent) = 1 : 1 under anaerobic conditions at 60 °C. Within 120 h, a gradual morphology transition of nanorod→nanowire→nanobelt can be observed with the disappearance of irregular structural fragments (Fig. 3a). Extending the self-assembly time to 120 h renders the formation of single-crystalline nanobelts with a length ≥1 mm (Fig. 2c), the width of ca. ~500 nm (Fig. 3a), and the thickness of ca. ~250 nm (Fig. 3b). The morphology of self-assembled :PDI2−/PDI2− nanobelt no longer changes after 120 h. Separated by centrifugation, obtained all-organic semiconductors remain stable in solvents, including methanol, ethanol, and water, or as a powder for more than half a year without further morphological and structural transitions (Supplementary Fig. 12).

Fig. 3: Structural characterizations of self-assembled :PDI2−/PDI2− organic semiconductor.
Fig. 3: Structural characterizations of self-assembled :PDI2−/PDI2− organic semiconductor.
Full size image

a SEM images to show morphology evolution of ·PDI2− during the self-assembly. b AFM images of :PDI2−/PDI2− samples after 24 h and 120 h self-assembly and corresponding line-scan profile to show the thickness of :PDI2−/PDI2− nanobelt. c Single-crystal SAED pattern of :PDI2−/PDI2− nanobelts (120 h). d A simulated arrangement model of :PDI2−/PDI2− crystal according to the SAED and XRD patterns. e XRD patterns and ESR spectra of :PDI2−/PDI2− with different self-assembly times within 120 h. f Profiles of the changing a-axis length and 2θ diffraction angle during the self-assembly were calculated according to the {100} peak position in XRD patterns.

Typically, the self-assembly of PDI derivatives occurs via the strong π-interaction between perylene rings, forming a vertical face-to-face planar stacking H-type configuration. In contrast, some polar group-terminated PDI molecules can follow a J-type arrangement in a rotational head-to-tail stacking model. N, N’-di(propanoic acid)-perylene-3, 4, 9, 10-tetracarboxylic diimide is well-reported that follows the H-type configuration28. However, albeit sharing an identical molecular framework with PDI, :PDI2−/PDI2− aggregates show a distinct J-type pattern (Supplementary Figs. 13, 14 and Supplementary Table 1) with redshifted visible-light absorption after the self-assembly due to π-stacking splitting (see below). Combined with the single-crystal selected area electron diffraction (SAED) pattern (Fig. 3c), a simulated crystal model of :PDI2−/PDI2− organic semiconductor can be visualized (Fig. 3d). Notably, after the self-assembly, the fluorescence emissions of the PDI2− quinoid isomer can still be observed (Supplementary Fig. 15), indicating that the tautomeric equilibrium between :PDI2− and PDI2− maintains in self-assembled crystals.

Moreover, after the self-assembly, the lattice parameter of ːPDI2−/PDI2− crystal evolves as the {100} peak shifts towards larger angles in 120 h (Fig. 3e). Such a shift epitomizes simultaneous features combining the disappearance of irregular fragments and the growth of self-assembled crystals, corresponding to a shortened apparent lattice length along the a-axis of the self-assembled crystal (Fig. 3f). Meanwhile, the ESR feature of ːPDI2− also shifts with the self-assembly time (Fig. 3e). After 96 h of self-assembly, a new splitting feature at g = 2.0044 and g = 2.0037 corresponding to the non-zero field transition of the spin-triplet state (STS) emerge, which implies a solid spin-orbit coupling within triplet ːPDI2− diradicals. Such a triplet spin-orbit coupling can occur on neighboring diradicals or between quinoids and diradicals, forming the triplet-singlet diradical-quinoid (D-Q) hybrid state. Since a shortened a-length from strain narrows the intermolecular distance between building blocks, the interaction between unpaired electrons can be enhanced, leading to an intermolecular triplet spin-orbit coupling within the patterned self-assembled ːPDI2−/PDI2− crystal. Moreover, a broad NIR absorption band peaking at ~700 nm emerged after the spin coupling, which broadened the visible-light absorption range of self-assembled ːPDI2−/PDI2− nanobelt close to the profile of the solar spectrum (Fig. 4c). Such an extra NIR absorbance corresponds to an unusual direct S0 → Tn transition that is typically spin-forbidden in organic molecular chromophore systems. Due to the forbidden nature of the immediate optical transition between the spin-zero ground (S0) state and the spin-one triplet (T1) levels, molecular triplet exciton is a dark state with no direct triplet absorption29,30. However, pioneering works that couple organic excimers with unpaired spins of heavy atoms such as Lanthanide enable a bright triplet NIR absorption through the triplet energy transfer21,22,23,24,25,29,30. In our system, the NIR absorption occurs in the interval of enhanced triplet spin-orbit coupling (72~96 h) (Fig. 5a), consistent with the feature of the direct triplet excitation of S0 → Tn transition on the D-Q hybrid. Such a direct S0 → T1 transition not only activates ‘dark’ triplet :PDI2− diradicals to the NIR absorbance, but also provides the opportunity to double the energy of these formed triplet excitons through a TTA upconversion for water-splitting reaction.

Fig. 4: Photocatalytic overall water-splitting over :PDI2−/PDI2− organic semiconductor catalysts.
Fig. 4: Photocatalytic overall water-splitting over :PDI2−/PDI2− organic semiconductor catalysts.
Full size image

a Comparisons of H2 and O2 productions from pure water on :PDI2−/PDI2− catalysts with different self-assembly times (0~120 h) under the white light illumination (≥300 nm, 1000 mW cm−2). Error bars (in standard deviation) were obtained by statistically repeating identical experimental results three times. b Time-profiles of H2 and O2 productions from pure water on PDI, :PDI2−/PDI2− without self-assembly, :PDI2−/PDI2− nanorod (24 h), and :PDI2−/PDI2− nanobelt (120 h) under identical AM 1.5 G simulated solar irradiation (1 sun, 100 mW cm-2) (performed once). c Wavelength-dependent AQYs for water splitting over :PDI2−/PDI2− nanorod (24 h) and :PDI2−/PDI2− nanobelt (120 h) along with corresponding UV-vis absorption spectra and the solar spectrum measured at 2 p.m. on June 13, 2022, at Jiangsu University (Zhenjiang, China). d GC-MS profiles of isotopically labeled gas products from water-splitting over :PDI2−/PDI2− nanobelt (120 h) in deuterium-labeled D2O and 18O-labeled H218O at light-off and light-on conditions, respectively. Raw data of the isotopically labeled see Supplementary Figs. 18–22. e Long-term recycling gases production performances over :PDI2−/PDI2− nanobelts (120 h) under white light (1000 mW cm-2) in 150 h with and without 15-min interval continuous vacuum pumping (performed once).

Fig. 5: Photoluminescence characterizations of :PDI2−/PDI2− organic semiconductors.
Fig. 5: Photoluminescence characterizations of :PDI2−/PDI2− organic semiconductors.
Full size image

a UV-vis absorption spectra of :PDI2−/PDI2− with different self-assembly times to show the direct S0 → Tn transition. b PL spectra (λex = 450 nm) of :PDI2−/PDI2− with different self-assembly times within 120 h. c Time-resolved transient PL emission decay (λex = 450 nm) on :PDI2−/PDI2− nanobelt (120 h) with λem = 540 nm, 585 nm, and 729 nm, respectively. d Upconversion PL spectra (λex = 808 nm) of :PDI2−/PDI2− with different self-assembly times (72~120 h). e PL emission (λex = 450 nm) of :PDI2−/PDI2− (after 84 h of self-assembly) quenched by water. f PL spectra (λex = 980 nm) of :PDI2−/PDI2− with different self-assembly times (72~120 h). No upconversion emission was observed.

Photocatalytic overall water-splitting

Figure 4a shows the quantified evolution rates of H2 and O2 from 50 mL water containing 0.05 g ːPDI2−/PDI2− catalysts with different self-assembly times under the constant white light illumination (Xe lamp, central spot intensity 1000 mW cm-2), in which H2 and O2 productions can be observed without introducing any metal-based cocatalysts. The discrepancy in the H2/O2 deviating from the ideal 2 :1 stoichiometric ratio is from the inevitable H2O2 production, which accounts for approximately 60 µmol g−1 h−1 (Supplementary Fig. 16 quantified by the I- titration method). The performance of ːPDI2−/PDI2− catalysts has been improved in order of magnitudes after the self-assembly. Notably, a leap of the performance (~2.5 times) can be observed on ːPDI2−/PDI2− catalysts between 72 h and 96 h of self-assembly (Fig. 4a), consistent with the interval of triplet spin-orbit coupling (Fig. 3e). Further extending the self-assembly time to 120 h shows the best hydrogen production rate (HPR). As controls, water-splitting experiments on PDI before and after self-assembly were performed with no H2 and O2 formation detected (Supplementary Fig. 17). Isotopically labeled experiments were further conducted to confirm that observed H2 and O2 productions were sorely from the photocatalytic water splitting rather than other effects (Fig. 4d, Supplementary Figs. 18–22). A turnover number (TON) > 19 was achieved within 150 h by performing the long-term reaction over the self-assembled ːPDI2−/PDI2− nanobelt catalyst, solidly verifying the occurrence of overall water splitting (Fig. 4e). Due to the intense visible-to-NIR absorption in the range of 450 nm~850 nm, self-assembled ːPDI2−/PDI2− nanobelt catalyst exhibits superior efficiencies under the AM 1.5 G simulated sunlight (1 sun, 100 mW cm-2) (Fig. 4b) and apparent quantum yields (AQY) in the visible-to-NIR region (Fig. 4c). The ːPDI2−/PDI2− nanobelt catalyst exhibits a record AQY of 1.96% at 550 nm and the average AQY exceeds 1.5% in the range of 550–850 nm, the most intense radiation band of the solar spectrum (Supplementary Tables 2–4, performed once). More importantly, with the λ = 850 nm excitation, an AQY of 0.95% was obtained. Such a single-band NIR-driven water-splitting is impossible with direct excitations by low-energy photons (<1.5 eV) due to the overpotential and thermodynamical requirement1,2,3,4. The STH efficiency of the ːPDI2−/PDI2− nanobelt catalyst was determined to be ~0.0806%, which is the first example of solar water splitting over all-organic materials. Our self-assembled ːPDI2−/PDI2− catalyst only stably works under continuous vacuum pumping to avoid the excessive accumulation of O2 (Fig. 4e), which can slowly corrode the material.

Mechanism analysis

We attribute the solar overall water-splitting efficiency over the self-assembled ːPDI2−/PDI2− catalyst to the following two aspects : i) accommodating the two-band excitation, i.e. spin-allowed SF and spin-forbidden S0 → T1 transition, to coherently utilize the full visible-to-NIR solar spectrum to produce triplet excitons for TTA-based upconversion; ii) decreased HER overpotential by hydrophilic diradical ːPDI2− moiety with highly delocalized unpaired electrons. We first investigate the excitation feature through the steady-state PL emissions of ːPDI2−/PDI2− solid samples with different self-assembly times. Under 450 nm excitation (Fig. 5b), ːPDI2−/PDI2− samples only exhibit singlet emissions of the quinoid moiety at 540 nm and 585 nm with self-assembly times of 24 h, 48 h and 72 h. However, with the increasing self-assembly times, ːPDI2−/PDI2− samples from 84 h to 120 h additionally show a broad and intense emission band transitioning from 654 nm to 729 nm on the original quinoid singlet emission feature (Fig. 5b), indicating a new S1 state in sharp contrast to that of the quinoid moiety. Such a broad PL emission appears on ːPDI2−/PDI2− samples after 84 h of self-assembly, coinciding with the spin coupling threshold (Fig. 3e). Due to the unchanged PL feature of the quinoid moiety and the non-luminescent property of ːPDI2− diradicals, we ascribe the newly formed PL emission to the TTA upconversion on the D-Q hybrid (Fig. 5b). Under 450 nm excitation, such a TTA upconversion are only available on well-assembled crystals, which we believe arise from the spin-allowed S0S1 → 1(TT) SF on the diradical chromophore and subsequent 1(TT) → T1 + T1 dissociation process to produce the final T1 state (diffusive in the D-Q hybrid). On the 120 h self-assembled ːPDI2−/PDI2− nanobelt, the quantum efficiency for the TTA upconversion emission (λem = 719 nm) was measured as 0.688%. Typically, two-photon or multi-photon upconversion processes exhibit nonlinear behaviors31. However, in our system, the triplet fusion upconversion (TFU) occurs immediately following SF, which differs from the conventional upconversion mechanisms. We further performed PL emission spectra measurements with varying light intensities (10 µJ cm−2 ~ 30 µJ cm−2) (Supplementary Fig. 23a). The observed linearity does not conform to the typical two-photon upconversion mode, showing a linear relationship (Supplementary Fig. 23b), which is consistent with our SF-TFU transition route. The SF and TFU processes are inversely related and interconnected, leading to a less pronounced nonlinear behavior. Similar phenomenon has been observed on other organic photovoltaic cells that employs a pentacene-based SF process with a sublinear dependence on excitation intensity, which is distinct from the typical nonlinear response in conventional multi-photon upconversion systems32. Notably, the lifetime of the upconversion emission remains unchanged with different incident light intensities (Supplementary Fig. 23c), suggesting that no saturation or intensity-induced quenching (such as exciton-exciton annihilation or energy pooling) occurs during this process. When same samples were measured under 808 nm excitation, an identical transitioning PL emission from 645 nm to 705 nm after 84 h of self-assembly was observed with singlet emissions of the quinoid moiety at 540 nm and 585 nm disappeared as expected (Fig. 5d). Since the S1 state of both ːPDI2− and PDI2− chromophores are above the 808 nm excitation energy (Fig. 5a), triplet excitons are no longer produced via SF, but from a direct S0 → T1 transition on the D-Q hybrid. This provides solid evidence for the self-assembly-induced triplet-fusion upconversion.

Furthermore, we conducted femtosecond transient absorption spectroscopy (fs-TAS) measurements on the :PDI2−/PDI2− samples before and after 120 h self-assembly. The 120 h self-assembled :PDI2−/PDI2− sample showed a rapid singlet fission (SF) process (occurring in 300 fs) and the emergence of a triplet excited state absorption (ESA) feature (around 770 nm) after SF upon λex = 540 nm excitation (Supplementary Fig. 24d–g), with the SF yield calculated to be approximately 42%. This observation contrasts with the behavior of the :PDI2−/PDI2− molecular precursor (Supplementary Fig. 24a–c). An identical triplet ESA at ~770 nm, resulting from the direct S0-T1 transition, was also observed under λex = 700 nm excitation (Supplementary Fig. 24h–j). Detailed data analysis and signal assignment was described in Supplementary Information. Our fs-TAS observations are consistent with our photoluminescence (PL) emission experiments and strongly support our proposed mechanism.

The transitioning redshift of the TTA-upconversion emission is due to the enhancing spin coupling between diradicals and quinoids with the increasing self-assembly times, similar to that in the rubrene-lanthanide system21. Furthermore, the upconversion emission of the 120 h self-assembled sample at 729 nm has a much longer lifetime (τ = 4.33 ns) than that at 540 nm (τ = 0.67 ns) and 585 nm (τ = 0.79 ns) (Fig. 5c). A longer emission decay time directly corresponds to a slower back-electron-transfer rate33, which facilitates the ns-scale interfacial water-splitting reaction. This new S1 state on the D-Q hybrid with a maximum 729 nm emission indicates that the energy of the parent T1 state is lower than 0.85 eV, which is below the 3O2 → 1O2 (0.94 eV) quenching energy and guarantees no interference of formed O234. More importantly, the observed TTA upconversion emission could be severely extinguished by water (Fig. 5e), while singlet emissions at 540 nm and 585 nm of quinoid moiety could not. This indicates that the new S1 excited state on the D-Q hybrid from TTA-upconversion is water-reactive, consistent with the unusual NIR-driven water-splitting efficiency on ːPDI2−/PDI2− nanobelt. We further conducted temperature-dependent PL measurements on the :PDI2−/PDI2− (120 h) sample over 100 K to 300 K (Supplementary Fig. 25a). Through the Arrhenius kinetic analysis, the binding energy (Eb) of *S1 singlet excitons after the TTA upconversion is determined as 21 ± 2 meV (Supplementary Fig. 25b), which is significantly lower than the binding energies typically observed in other organic semiconductors. This value is less than the Eb of the S1 state associated with the 540/580 nm emission and the KBT value (~26 meV) at room temperature, suggesting that TTA-resulted *S1 singlet excitons are highly active and readily dissociate into electrons and holes to participate in reactions, which is consistent with the fluorescence quenching in water (Fig. 5e).The self-assembly of such spin-coupled ːPDI2−/PDI2− chromophore renders intense adsorption features in the NIR region (700 nm~1100 nm) (Fig. 5a), matching well with the direct S0 → Tn transition feature with 0-0, 0-1, and 0-2 vibrational states21,22,23,24,25. However, with the 980 nm excitation (~lowest 0-0 vibrational level), the upconversion emission on samples did not appear (Fig. 5f), implying that the TTA-upconversion requires photon energies sufficient for the S0 → T1 transition.

Besides the two-band excitation to make the full use of the visible-to-NIR solar spectrum, breaking through the HER bottleneck with even a much lower S1 state level on the D-Q hybrid than that either on ːPDI2− or PDI2− monomer chromophore is another important issue. A radical-mediated electron transfer (ET) pathway is reported to hold much lower reaction overpotentials, allowing reactions that are impossible under the non-radical pathway to occur35,36. In our system, the interfacial ET of HER occurs on the ordered ːPDI2−/PDI2− surface with plenty of diradical unpaired electrons, which makes the H+ → H2 reduction easier along the favorable diradical-mediated pathway. In addition, the enhanced hydrophilicity of ːPDI2−/PDI2− nanobelt may also play an important role. Unlike the generally inert covalent H-coordination of organic surface, ːPDI2−/PDI2− nanobelt has an ionic diradical dipole with enhanced affinity with protons to reach the charge balance, forming a hydrophilic surface with better wettability (Supplementary Fig. 26), which is essential for the adsorption and dissociation of H2O molecules at the water/catalyst interface. Moreover, the small Eb of TFU-resulted *S1 excitons (21 ± 2 meV) makes the reduced :PDI2− diradicals exhibit properties more akin to inorganic materials, contributing to the breaking through of HER bottleneck without noble-/transition-metal components.

Our work is the proof-of-concept example of a metal-free organic catalyst capable of vis-to-NIR-driven overall water-splitting. Furthermore, such a self-assembly strategy of open-shell/closed-shell resonant structure applies to multiple derivatives of the PDI family (five derivatives verified) (Supplementary Fig. 27). This provides a path to design a series of brand-new photo-response organic semiconductors for photocatalysis, solar cells, sensors and light emitting materials.

Methods

Chemicals

Perylene-3, 4, 9, 10-tetracarboxylic dianhydride (≥92.0%), β-Alanine (99%), hydrochloric acid (HCl, 37%), hydrazine hydrate (N2H4·H2O, >85%), imidazole (99%), 1, 10-phenanthroline (99%), Fe(NO3)3 (99.99%), sodium acetate (99%), tetrabutylammonium tetrafluoroborate (≥98%), dimethyl sulfoxide (AR, >99%) and acetic acid (GR, 99.8%) were all purchased from Aladdin Biochemical Technology Co., Ltd. China. Deuterium oxide (D2O, 99 atom % D) and heavy oxygen water (H218O, 97 atom % 18O) were obtained from Sigma-Aldrich Company Ltd. China.

Preparation of PDI

In a typical preparation procedure, 1.376 g (3.507 mmol) of perylene-3,4,9,10-tetracarboxylic dianhydride, 2.5 g (28.06 mmol) of β-Alanine, 18 g of imidazole were placed in a mortar and evenly grounded. Immediately transferred to a porcelain boat, heated in a tube furnace at 100 °C for four hours under the protection of Nitrogen. The resulting solid product was then uniformly dispersed in 100 mL of ethanol with the addition of 300 mL of 2 M HCl. After magnetically stirring overnight, the resulting mixture was centrifuged and washed with deionized water until the solution was neutral (pH ~7). Obtained solid was collected and dried at 60 °C in the vacuum for further use.

Preparation of :PDI2−/PDI2− precursor

In a typical preparation procedure, 1.0 g of freshly prepared PDI was first dissolved uniformly in 100 mL hydrazine hydrate (>85%). The reduction reaction should last for more than 4 h. Then evaporating the solvent with the protection of Nitrogen flow. Obtained solid was collected and washed with ethanol, and dried on a 60 °C hot stage until constant weight (in a Nitrogen glove box). The obtained product is a mixture of open-shell diradical and the tautomeric closed-shell quinoid isomer with a ratio of 74% to 26%, denoted :PDI2−/PDI2−, which can be stably stored under the anaerobic condition.

Self-assembly of ːPDI2−/PDI2− nanobelt

In a typical preparation procedure, 10 mg ːPDI2−/PDI2− precursor was dissolved uniformly in 5 mL of deionized water (good solvent), then 5 mL ethanol (poor solvent) was added. The mixed solution was transferred into a 10 mL vial. The vials were sealed and placed on a 60 °C hot stage for 0–120 h. The samples with different self-assembly times were centrifuged and washed with water and dried until constant weight. The above operations were carried out in the Nitrogen glove box.

Water-splitting performance test

The photocatalytic water-splitting over different catalysts was carried out in an outer irradiation-type photoreactor (quartz glass) connected to a closed gas-circulation system. In a typical procedure, 0.05 g of the photocatalysts were dispersed by a magnetic stirrer in 50 mL of deionized water. The catalyst was transferred from the glove box by injection sampling to avoid oxidation of the catalyst. The suspension is thoroughly degassed to remove air and irradiated using a 300 W Xe-lamp (1000 mW cm−2, CEL-HXF300-T3, Beijing China Education Au-light Technology Co. Ltd, China). The photocatalytic H2 and O2 evolution rate was analyzed using an online GC-7920 gas chromatograph (GC, TCD detector, 5 A˚ molecular sieve columns and Ar2 carrier). Magnetic stirring (400 rpm) was used during the water splitting experiment to ensure homogeneity of the suspension and to eliminate sedimentation. With simulated AM1.5 G sunlight irradiation, a solar simulator (100 mW cm−2, Xenon Light Source CEl-S500-T5, Beijing China Education Au-light Technology Co. Ltd, China) was employed as the light source.

Isotopically labeled GC-MS measurement

Preparation of isotopically labeled samples

In the Nitrogen glove box, 20 mg ːPDI2−/PDI2− nanobelts was dispersed uniformly in 10 mL of D2O (H218O or H2O) and transferred into a 50 mL volume vial. The gas (5 mL) was collected by the injection needle with a throttle valve into sealed gas bag, denoted as Light-off sample. The vials were sealed and irradiated using a 300 W Xe-lamp (1000 mW cm−2, CEL-HXF300-T3, Beijing China Education Au-light Technology Co. Ltd, China) for 2 h. The reaction gas (5 mL) was collected by the injection needle with a throttle valve into sealed gas bag, denoted as Light-on sample. The Light-off samples and Light-on samples were collected from the same vial.

Using agilent CP-Molsieve 5A as a column for isotope-labeled samples analysis

The 2 mL gas samples were collected and manually injected by gas-tight syringes (the FTFE luer lok (PN5190-1534), 0~2.5 mL) and then analyzed by gas chromatography-mass spectrometry (8890-5977B GC-MS instrument, Agilent Technologies, USA) equipped with an Agilent CP-Molsieve 5A column (25 m × 250 μm × 30 μm, Agilent Technologies, USA) in GC-MS. Helium was used as carrier gas. The column was maintained at 100 °C for 5 min, and the flow of the carrier was 1.5 mL min−1. The temperatures of the injector and MS source were set to be 100 °C and 230 °C, respectively. The ionization potentials used for analysis were set to 20 eV and 70 eV, respectively.

In addition, we repeated the isotopically labeled GC-MS measurement using headspace vial (sample through drainage method) under above mentioned chromatographic conditions. The ionization potentials used for analysis were set to 18 eV and 70 eV, respectively.

Fe(III)-1, 10-Phenanthroline spectrometric titration measurements

The reductive electron concentration on ːPDI2−/PDI2− precursor was measured by a Fe(III)-1, 10-phenanthroline titration spectrometric method. 1, 10-phenanthroline spectrometric measurement is a simple and widely used method for measuring Fe(II) ions. Here, we use the Fe(III) solution to titrate the unpaired electrons on ːPDI2−/PDI2− precursor that quantitatively lead to the production of Fe(II) ion, then used 1, 10-phenanthroline to measure the concentration of Fe(II) ions. Thus, we can quantitatively obtain the concentration of trapped electrons. The concentration of Fe(NO3)3 solutions employed in this measurement is 10−3 M. 0.2% 1, 10-phenanthroline aqueous solution and pH = 4.6 HAc-NaAc buffer solution were previously prepared for use. The pH=4.6 HAc-NaAc buffer solution was prepared by dissolving 135 g sodium acetate and 120 mL acetic acid into 500 mL water solution. Before the titration, the Fe(NO3)3 solution is purged by nitrogen for 30 min to remove oxygen. The titration was conducted in the glove box. In a typical procedure, 2.5 mL ːPDI2−/PDI2− aqueous solution was taken in the glove box and mixed with 2.5 mL Fe(NO3)3 solution. 5 min later, the resulting mixed solution was taken out to conduct the spectrometric measurement. The spectrometric measurement was conducted in the air. 1.5 mL supernatant taken from the glove box was mixed with 1.5 mL pH = 4.6 HAc-NaAc buffer solution, and then added with 1 mL 0.2% 1, 10-phenanthroline water solution to obtain a red solution. Waiting 5 min to obtain a stable state, the resulting solution was transferred in a quartz cuvette and measured on a Hitachi U3900 spectroscopy. Obtained absorbance value was compared with the standard fitting line to obtain a certain Fe(II) concentration. The background data was collected by the same method except replacing 1, 10-phenanthroline solution with deionized water to eliminate the influence of PDI absorbance.

Characterization devices

The crystalline structure of the resultant products was characterized by X-ray diffraction (XRD) using an X’Pert-ProMPD (Holand) D/max-γA X-ray diffractometer with Cu Kα radiation (λ = 0.154178 nm) at a scan rate of 5° min−1. The near-ambient pressure x-ray photoelectron spectroscopy (NAP-XPS) experiments were conducted on a laboratory-based SPECS near-ambient pressure XPS system. X-ray photoelectron spectroscopy (XPS) measurements were performed using a Nexsa instrument (Thermo Fisher Scientific) with Al Kα radiation (hυ = 1486.6 eV) as the excitation source. Scanning electron microscopy (SEM) images were performed by a FEI-quanta 200 scanning electron microscope with an acceleration voltage of 20 kV. The electron paramagnetic resonance (EPR) analysis was performed on a Bruker EPR JES-FA200 spectrometer. Transient fluorescence decay spectra were characterized on an Edinburgh FLS1000 fluorescence spectrometer. The ultraviolet-visible diffuse reflectance spectroscopy (UV-vis DRS) was recorded on a Shimadzu UV-3600 spectrometer. Atomic force microscopy (AFM) was performed by means of a Veeco DI Nanoscope Multi Mode V system. The isotopically labeled experiments were carried out on a gas chromatography-mass spectrometry (GC-MS, Agilent 7890B-5977B).

AQY calculations

The AQYs at different wavelengths (λ = 365–850 nm) for ːPDI2−/PDI2− catalysts were determined by taking catalyst solution (0.05 g catalyst and 50 mL deionized water) irradiated by a 300 W Xe lamp applying different band-pass filters for 5 h. The average intensity of irradiation was determined by a CEL-NP2000-10A spectroradiometer (Beijing China Education Au-light Technology Co. Ltd, China) and the irradiation area was 4 cm2 (2 cm × 2 cm). For example, when λ = 550 nm, the average intensity of irradiation was determined to be 269.8 mW cm-2. The amount of H2 generated in 5 h was 87.5 μmol. The number of incident photons (N) is as calculated by the following equation:

$${{{{{\rm{N}}}}}}=\frac{E{{{{{\rm{\lambda }}}}}}}{{hc}}=\frac{269.8 \times 4 \times {10}^{-3} \times 5 \times 3600 \times 550 \times {10}^{-9}}{6.626 \times {10}^{-34} \times 3 \times {10}^{8}}=0.5375 \times {10}^{22}$$
(1)

So,

$${{{{{\rm{AQY}}}}}}= \frac{2\times {{{{{\rm{the}}}}}} \; {{{{{\rm{number}}}}}} \; {{{{{\rm{of}}}}}} \; {{{{{\rm{evolved}}}}}}\,{{{{{{\rm{H}}}}}}}_{2}\,{{{{{\rm{molecules}}}}}}}{{{{{{\rm{N}}}}}}} \times 100 \% \\= \frac{2 \times 6.02 \times {10}^{23} \times 87.5 \times {10}^{-6}}{0.5375 \times {10}^{22}} \times 100 \%=1.96 \%$$
(2)

STH calculations

The STH efficiency was evaluated using AM 1.5 G solar simulator (1 sun) as the light source with an irradiation area of 2.25 cm2 (1.5 cm × 1.5 cm). With ːPDI2−/PDI2− nanobelt (120 h) as the catalyst (0.05 g catalyst in 50 mL water), the H2 generated in 5 h was 275.55 μmol (RH2 = 7.654 × 10−7 mmol s−1).

So,

$${{{{{\rm{STH}}}}}}=\frac{{\Delta G}_{r} \times {R}_{H2}}{{P}_{{Sun}} \times S} \times 100 \%=\frac{2.37 \times {10}^{3} \times {R}_{H2}}{S} \times 100 \% \\ {{{{{\rm{STH}}}}}}=\frac{2.37 \times {10}^{3} \times 7.654 \times {10}^{-7}}{2.25} \times 100 \%=0.0806\%$$
(3)

Here, \({R}_{{H}_{2}}\): The H2 evolution rate (mmol s−1);

\({\Delta G}_{r}\): The reaction Gibbs energy of water splitting (J mol−1);

\({P}_{{Sun}}\): Light power density of the AM 1.5 G standard solar spectrum (100 mW cm−2);

\({{{{{\rm{S}}}}}}\): The irradiation area.