Introduction

Photocatalytic overall water splitting (OWS) using low-cost and eco-friendly polymeric semiconductor photocatalysts holds great promise for advancing hydrogen-based energy society1,2,3,4,5. As a representative of high-crystalline polymers, poly (triazine imide) (PTI) has attracted intensive attention due to its considerable potential in OWS6,7,8,9,10,11. However, despite increasing attempts to enhance the photocatalytic properties of PTI through structure and composition modulation12,13,14,15, its OWS performance still lags behind that of inorganic semiconductors like SrTiO316,17,18,19,20,21.

When tracing the root cause of this disparity, it turns out that the indispensable properties for highly efficient metal-element contained photocatalysts should include: (1) efficient transport and anisotropic separation of photo-induced charge carriers in the bulk22,23,24,25,26,27,28; (2) spatially separated distribution of reduction and oxidation sites29,30,31,32,33,34,35,36. However, these characteristics seem to be inherent shortcomings that polymeric semiconductors naturally lack. In fact, as a representative conjugated organic polymer with high crystallinity, the reported smallest exciton binding energy is still as high as 43 meV13. Furthermore, the overlap of reduction and oxidation sites on its surfaces means that photoexcited charge carriers in PTI predominantly exist as excitons8, which greatly limits photocatalytic performance due to the low efficiency of charge separation and carrier migration. High exciton binding energy (Eb) remains a critical bottleneck for organic semiconductor photocatalysts. Despite continuous efforts to mitigate this issue, existing strategies for reducing Eb in photoactive organic semiconductors primarily include: (i) incorporating donor-acceptor (D-A) polarized motifs into covalent organic frameworks (COFs)37,38,39; (ii) constructing heterostructures or introducing dopants to lower Eb in poly(heptazine imide) (PHI)14,40; and (iii) enhancing crystallinity or introducing heteroatom doping in graphitic carbon nitride (g-C3N4)41,42. Despite these advancements, none of these modified organic polymers have demonstrated viable performance for overall photocatalytic water splitting. Notably, poly(triazine imide) (PTI) stands as a rare exception capable of achieving complete water dissociation, yet its reported Eb values (far exceeding the thermal energy at room temperature, 25.7 meV @ 25 °C) remain unaddressed by targeted strategies. The lack of effective approaches to reduce its inherently high Eb severely impedes further improvements in the OWS performance of PTI.

In principle, delocalized electronic states and structural anisotropy of semiconductor particles are two essential prerequisites for exciton dissociation27,43,44, as they provide the migration channels and driving forces for exciton separation, respectively. On one hand, for the PTI atomic planes, the triazine framework, composed of C-N covalent bonds, causes all energy states to be strongly localized around the triazine rings, which may hinder sufficient exciton dissociation8. On the other hand, the highly symmetric nature of the hexagonal structure in PTI typically makes it challenging to achieve anisotropic carrier transport preference45. To date, while common approaches involving alkaline metal eutectic salts with similar lattice structures, such as LiCl/KCl, NaCl/LiCl, and NaCl/KCl/LiCl, have enhanced conjugation and reduced defects in the material, thereby improving the OWS performance of PTI to some extent11,12,13, significantly reducing its exciton binding energy remains a challenge. To date, no effective strategy has been developed for the efficient dissociation of excitons into free charges in PTI photocatalysts. This limitation hinders the material’s potential in photocatalytic OWS. Therefore, achieving effective exciton dissociation is key to advancing PTI-related research.

Herein, a lattice engineering strategy was employed by altering the LiCl/KCl eutectic salt for the conventional synthesis of PTI (PTI-LiK, hereinafter) to LiCl/CaCl2, thus resulting in PTI nanoplates with excitons fully dissociated to free charges (PTI-LiCa, hereinafter). Experimentally, in-plane lattice contraction in PTI has been observed, resulting in the transition of excitons to free charges, which was confirmed by temperature-dependent photoluminescence (TD-PL) spectroscopy, surface photovoltage spectra (SPV) and transient photovoltage spectra (TPV), etc. techniques. Furthermore, the in-plane lattice contraction boosted the delocalization of the energy states, and doping of Ca2+ initiated the anisotropic distribution of delocalized highest occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) through theoretical calculation, thus triggering the spontaneous dissociation of the photo-induced excitons into free charges in PTI. At last, the chemically traceable anisotropic separation of photo-induced electrons and holes has been observed in PTI. In addition, the enhanced overall water-splitting (OWS) activity of PTI-LiCa compared to PTI-LiK highlighted the qualitative improvements resulting from the complete dissociation of excitons in such promising polymeric photocatalysts.

Results

Synthesis of the PTI nanoplates with lattice contraction and Ca dopants

As depicted in Fig. 1a, in contrast to the (100) atom plane of cubic KCl, which can be served as the nucleation and growth template of PTI-LiK hexagonal prisms, the PTI-LiCa sample exhibits nanoplates when the (\(\bar{1}\)12) planes of orthogonal CaCl2 act as the growth template. The morphology of PTI-LiCa shown in Fig. 1b, c, and Supplementary Figs. 1–3, suggests that the hexagonal bases of nanoplates can be attributed to the {001} facets of PTI. Specifically, the diagonal length (from one corner to its opposite corner) of a typical PTI-LiCa hexagonal nanoplate is about 60 nm, and its thickness is about 20 nm. In sharp contrast, the length of the PTI-LiK prims is as long as 120 nm, with their hexagonal bases showing similar sizes to that of PTI-LiCa, as shown in Supplementary Figs. 4 and 5, the aspect ratios of which are close to those reported. As a sign for the lattice contraction, the interplanar spacing in Fig. 1c corresponding to {10\(\bar{1}\)0} facets of PTI-LiCa were measured to be 7.27 Å, which is slightly smaller than that of PTI-LiK (7.33 Å) (see Supplementary Fig. 4b)12,13.

Fig. 1: Morphology, structure, and composition of PTI-LiCa.
Fig. 1: Morphology, structure, and composition of PTI-LiCa.
Full size image

a The synthesis scheme diagram of PTI nanoplates with in-plane lattice contraction on CaCl2 (\(\bar{1}\)12) as the growth template. b, c SEM, HRTEM images of PTI-LiCa. d XRD patterns of PTI-LiCa compared with PTI-LiK. e TEM image of PTI-LiCa grown on CaCl2 (\(\bar{1}\)12) collected from the eutectic salt mixture. f The root cause for the in-plane lattice contraction of PTI-LiCa grown on CaCl2 (\(\bar{1}\)12) facets based on the LCMP analysis. g The XANES of the Ca K-edge of PTI-LiCa with that of CaCl2 and CaO as the counterparts. The inset table is the EXAFS fitting results of PTI-LiCa. h Comparison of PTI structures with and without Ca doping. Source data are provided as a Source Data file.

Such lattice contraction was further confirmed by the comparison of X-ray diffraction patterns (XRD) of two PTI samples given in Fig. 1d and Supplementary Fig. 6. Notable XRD peak position shifts in PTI-LiCa could be observed compared with PTI-LiK. The 2θ values of the peaks corresponding to intralayer facets like (100), (110), (200), (210), (220) and (320) increased, while those of (002), (102), and (004) decreased, providing direct evidence for the intralayer lattice contraction as well as the interlayer enlargement in PTI-LiCa. Specifically, the 0.1o decrease in the 2θ value meant the lattice contraction by nearly 0.059 Å or 0.8%. Based on the Rietveld refinement via the General Structure Analysis System (GSAS) in Supplementary Fig. 7, compared with PTI-LiK, the in-plane lattice parameters (a and b-axis) of PTI-LiCa decreased from 8.46333 to 8.44959 Å, while the c-axis lattice parameter increased from 6.71196 Å to 6.78191 Å, showing an intralayer compressive strain of − 0.16% and an interlayer space enlargement of 1.04%, respectively. The structural difference between CaCl2 (orthorhombic phase) and KCl (cubic phase) in the eutectic salts was believed to cause the intralayer lattice contraction in PTI-LiCa. The high-resolution transmission electron microscopy (HR-TEM) image in Fig. 1e shows the lattice fringes corresponding to CaCl2 (\(\bar{1}\)12) and dark contrast hexagons attributed to the PTI-LiCa hexagonal nanoplates, suggesting the role of CaCl2 (\(\bar{1}\)12) facet as the growth substrate for the (001) planes of PTI-LiCa, like KCl (200) facets for PTI-LiK hexagonal prisms15. Based on the analysis via the Least Common Multiple Principle (LCMP), the PTI-LiK (001) planes grown on the KCl (001) facets showed only − 0.6% mismatch (Supplementary Fig. 8), while the growth of PTI-LiCa on CaCl2 (\(\bar{1}\)12) would suffer a mismatch as large as − 2.28% (Fig. 1f). More details of the LCMP analysis can be found in Supplementary Materials. As shown in Supplementary Fig. 9a–c, the solid-state 13C NMR, the Fourier transform infrared spectra (FTIR), and the Raman spectra all proved that the PTI-LiCa sample maintains the structural feature of PTI12. And the electron paramagnetic resonance (EPR) spectra showed the lower intensity of PTI-LiCa than that of PTI-LiK, as shown in Supplementary Fig. 9d, which means the lower density of nitrogen defects in PTI-LiCa.

It is worth noting that the Ca2+ doping was realized as a consequence of the substitution of CaCl2 to KCl during the synthesis route of PTI. Initially, the energy-dispersive X-ray spectroscopy mapping images (EDS, Supplementary Figs. 10 and 11) and X-ray photoelectron spectroscopy spectra (XPS, Supplementary Fig. 12) were performed to investigate the surface distribution of Ca dopants. According to the deconvolution results shown in Supplementary Fig. 12, the XPS binding energies for C 1 s, N 1 s, and Cl 2p in PTI-LiCa compared to those of PTI-LiK showed systematic positive shifts, which implies that Ca2+ formed strong bonds with both triazine framework and chloride anions. Furthermore, XPS depth profiles (Supplementary Fig. 13), Inductively Coupled Plasma Optical Emission Spectrometer (ICP-OES), and Combustion Elemental Analysis (EA) were carried out to investigate the Ca dopants in the bulk. These results demonstrated that the Ca dopant was uniformly distributed both on the surface and within the bulk of PTI-LiCa. The elemental compositions are listed in Supplementary Table 1–3. The homogeneous distribution of Ca dopants in PTI-LiCa was further validated via time-of-flight secondary ion mass spectrometry (TOF-SIMS) depth profiling. As shown in Supplementary Fig. 14, the parallel evolution of Ca and Li signals with sputtering depth confirms the uniform calcium incorporation throughout the bulk phase. To obtain the coordination environment of Ca in PTI-LiCa, a Ca K-edge X-ray Absorption Fine Structure (XAFS) test was conducted46,47. The rising pre-edge feature indicates coordination between Ca and the triazine framework of PTI (Fig. 1g). The Extended X-ray Absorption Fine Structure (EXAFS) spectra and the fitting results (Supplementary Fig. 15) further confirmed that Ca in PTI-LiCa coordinated with three N atoms and one Cl atom (inset in Fig. 1g)48,49,50. The experimentally-obtained coordination configuration of Ca2+ with both triazine framework and chloride anion was well-validated and consistently reflected in the theoretical model, as shown in Supplementary Fig. 16. Notably, although the doping site of Ca deviates from the in-plane position and tends to the gallery of the planes, it still only bonds with the triazine ring of a single atomic plane. Comparison of PTI structures with and without Ca doping is presented in Fig. 1h. In addition, DFT simulation results on the variation of system energy as a function of distance along the c-axis in the two PTI model (with and without Ca²⁺ insertion) clearly demonstrate that PTI-LiCa exhibits a higher tendency for discontinuous growth along the c-direction (Supplementary Fig. 17). Further details and supporting evidence are provided in the Supplementary Information.

Enhanced photocatalytic OWS activity in PTI-LiCa

Previous reports have demonstrated the indispensable role of the {10\(\bar{1}\)0} in enabling photocatalysis8, which indicates that a higher proportion of this facet corresponds to enhanced photocatalytic performance. Based on the statistical results of the morphology images of the PTI-LiCa and PTI-LiK particles, the percentages of {10\(\bar{1}\)0} facets of PTI-LiK prisms and PTI-LiCa nanoplates were 79.5% and 44.1%, respectively (Fig. 2a, inset). During the photocatalytic overall water-splitting (OWS) with CoOx oxygen evolution cocatalyst and Pt@Cr2O3 hydrogen evolution cocatalysts, the H2 evolution rate of PTI-LiCa is about 3.36 times higher than that of PTI-LiK. The H2 and O2 evolution rates on PTI-LiCa were determined at 156.0 µmol/h/100 mg and 72.5 µmol/h/100 mg, respectively (Fig. 2a). For PTI-LiK, the average H2 and O2 evolution rates were 46.4 µmol/h/100 mg and 18.4 µmol/h/100 mg, respectively. In combination with the specific surface areas derived from the BET method (Supplementary Fig. 18, 65.89 m2/g for PTI-LiK and 80.86 m2/g for PTI-LiCa), the specific surface area of {10\(\bar{1}\)0} facets in PTI-LiK and PTI-LiCa were calculated to be 52.38 m2/g and 35.65 m2/g, respectively. The OWS H2 evolution rate normalized to the specific area of {10\(\bar{1}\)0} facets in PTI-LiCa was 43.76 µmol/h/m2, 4.94 times that of PTI-LiK (8.86 µmol/h/m2). To exclude the potential of residual CaCl2 impact on catalytic performance, controlled amounts of CaCl2 were added to the PTI-LiCa reaction system. As shown in Supplementary Fig. 19, the results confirm that CaCl2 does not interfere with the intrinsic catalytic advantage of PTI-LiCa over PTI-LiK.

Fig. 2: Activity and stability of PTI-LiCa in photocatalytic overall water splitting.
Fig. 2: Activity and stability of PTI-LiCa in photocatalytic overall water splitting.
Full size image

a The quantitative analysis of the exposure ratios of the (10\(\bar{1}\)0) facets of PTI-LiK and PTI-LiCa particles. And the comparison of the OWS performance of PTI-LiK and PTI-LiCa under optimized conditions, with Xe-lamp irradiation. b The HER, OER and ORR performances of PTI-LiK and PTI-LiCa as electrocatalysts. c Comparison of the Pt 4 f XPS high-resolution spectra of the PTI-LiK and PTI-LiCa deposited with cocatalysts collected after the OWS tests. d Photocatalytic stability test of PTI-LiCa. (All photocatalytic water splitting tests were taken in an automatic reaction system: Perfectlight Sci&Tech, Labsolar-6A). Source data are provided as a Source Data file.

To understand the origin of the photocatalytic OWS activity enhancement of PTI-LiCa compared with PTI-LiK, the electrocatalytic features of the two samples associated with water splitting were investigated. Note the influence of the samples’ bulk conductivity has been well excluded by adding conductive carbon additives in our electrocatalytic approach (Supplementary Fig. 20). As shown in Fig. 2b, both the higher onset potentials and the lower currents of PTI-LiCa during the hydrogen evolution reaction (HER) and oxygen evolution reaction (OER) indicated minimal contribution from more active surfaces. In addition, the higher peak current density observed in the oxygen reduction reaction (ORR) suggests that the surface of PTI-LiCa is less effective in suppressing the back reaction of H2 and O2 during the OWS.

Pt/Cr2O3 was loaded on PTI-LiCa as the cocatalysts for hydrogen evolution for its advantage over Rh/Cr2O3, and Pd/Cr2O3 in the preliminary experiments (Supplementary Fig. 21). To elucidate the specific roles of cocatalysts in enhancing charge transfer and OWS efficiency, we conducted operando FT-IR spectroscopy on PTI-LiK and PTI-LiCa both with and without cocatalysts, as shown in Supplementary Figs. 22 and 23. It is noted that cocatalysts-loaded samples exhibited dramatically attenuated peroxide signal growth, confirming that CoOx/Pt@Cr2O3 promotes efficient O2 evolution rather than intermediate trapping. Further XPS analysis of Pt cocatalysts of the samples after the OWS tests (Fig. 2c) shows that the proportion of Pt0 in the PTI-LiK-based sample is lower than that in the PTI-LiCa-based sample (7.9% vs. 12.9%), but the proportion of PtIV in the former is higher than that in the latter (43.0% vs. 30.2%). This indicated a higher enrichment concentration of photo-induced electrons on the reductive sites of PTI-LiCa compared to PTI-LiK. To rule out the potential contribution of the higher Pt0 faction in the Pt/Cr2O3 deposited on PTI-LiCa on its OWS performance, we also tested the OWS performance comparison of PTI-LiK and PTI-LiCa loaded with 1 wt% Pt via impregnation (Supplementary Fig. 24), in which PTI-LiCa maintained the same OWS activity advantage over PTI-LiK. Therefore, the improved separation and transport properties of the photo-induced charges shall be the primary factor contributing to the enhanced OWS performances of PTI-LiCa.

Instead of summing the gas evolution values after evacuation at each sampling interval8, the gas evolution amounts in our OWS measurements were recorded through continuous accumulation. Charges that fail to efficiently separate, migrate, and participate in the reaction inevitably lead to severe recombination of the photocatalysts. The enhanced photo-induced charge separation and transport properties of PTI-LiCa were further evidenced by its long-term stability in OWS. In the long-term test lasting for 18 h, the OWS activity of PTI-LiCa tended to stabilize in the following 6 cycles (Fig. 2d). The wavelength-dependent apparent quantum yield in OWS of PTI-LiCa was calculated to be 5.1%, 4.79%, 1.14% and 0.09% at 350, 365, 380 and 405 nm, respectively (Supplementary Fig. 25). The performance of PTI-LiCa remained essentially stable in the 18-hour, three-hour-per-cycle test. In addition, PTI-LiCa consistently maintained its performance advantage over PTI-LiK in the long-term test of up to 100 hours (Supplementary Fig. 26).

Spontaneous exciton dissociation into free charges in PTI-LiCa

To explore the factors contributing to the improved performance, we initially investigated the photophysical properties of the material. The combination of UV-visible absorption spectra, Mott-Schottky plots and XPS valence spectra confirmed the basically close energy band structure of PTI-LiCa to PTI-LiK, with only the donor defects decreasing by two orders (Supplementary Fig. 27). The exciton binding energy determined from temperature-dependent photoluminescence spectra (TD-PL) (Fig. 3a) is 48.2 meV for PTI-LiK, consistent with the previous report8. In contrast, the exciton binding energy of PTI-LiCa was found to be as low as 15.4 meV (Fig. 3b), far below the thermal disturbance energy at room temperature (25.7 meV), which clearly suggests that the photogenerated charge carriers in PTI-LiCa are more likely to spontaneously dissociate into free charge carriers rather than remain as excitons.

Fig. 3: Evidence for the photo-induced free charges in PTI-LiCa.
Fig. 3: Evidence for the photo-induced free charges in PTI-LiCa.
Full size image

a, b Temperature-dependent photoluminescence (TD-PL) spectra from 10 K to 300 K of PTI-LiK and PTI-LiCa. c Photoconductivity dynamics comparison between PTI-LiK and PTI-LiCa. d, f 2D contour plots of femtosecond transient absorption spectra (fs-TAS) illustrating the photocatalytic dynamic process of PTI-LiK and PTI-LiCa. e, g fs-TAS spectra of PTI-LiK and PTI-LiCa, of which the time scale varies from femtoseconds to picoseconds. h Schematic diagram of exciton dissociation in the bulk of PTI-LiK and PTI-LiCa. i Comparison of reported exciton binding energy of typical polymer materials in recent years. Source data are provided as a Source Data file.

THz spectroscopy is particularly effective for distinguishing free carriers from excitons due to their distinct optical responses in the THz regime. Free carriers exhibit strong THz absorption, resulting in a complex conductivity dominated by its real part, characterized by e.g., a Drude-like response. In contrast, excitons can resonantly absorb THz photons, promoting transitions from the ground to higher excitonic states, which manifests as Lorentzian features. In systems with strong excitonic character, where the excitonic resonance lies far beyond the bandwidth of our spectrometer (~ 2 THz, or ~ 8 meV), excitons appear as a near-zero real conductivity and a finite imaginary component. To probe the nature and dynamics of photogenerated carriers, samples were excited with 400 nm laser pulses (~ 50 fs duration), and the resulting charge states were analyzed using THz probe pulses.

Figure 3c presents the complex photoconductivity spectra for both PTI-LiCa and PTI-LiK. Upon above-bandgap excitation, electron-hole pairs are generated in both materials, leading to an initial increase in the real component of the photoconductivity. Remarkably, in PTI-LiK, this transient free carrier population rapidly converts into nearly 100% excitons on a sub-picosecond timescale. The resulting conductivity spectrum shows near-zero real and finite imaginary components, indicating the dominance of bound excitonic states. This observation implies a high exciton binding energy in PTI-LiK, well above the thermal energy at room temperature (~ 26 meV), and is fully consistent with our combined spectroscopic studies, including PL, SPV, and TPV measurements. In contrast, PTI-LiCa maintains a strong free carrier response, evidenced by a finite real conductivity component. This suggests that the exciton binding energy in PTI-LiCa is sufficiently small that a substantial fraction of carriers remains free at room temperature.

To gain deeper insight into the enhanced charge transport properties experimentally, conductivity spectra derived from time-domain THz spectroscopy of PTI-LiK and PTI-LiCa films (with identical thickness of ~ 1 μm) were obtained in the dark, i.e., without photoexcitation, as shown in Supplementary Fig. 28. Notably, PTI-LiCa exhibits a higher real component of conductivity compared to PTI-LiK. In addition, PTI-LiCa demonstrates stronger frequency-dependent dispersion in the conductivity spectrum, indicating a longer momentum scattering time relative to PTI-LiK. Using the widely adopted Drude–Smith model (see the fitting in Supplementary Fig. 28), we infer carrier scattering times of approximately 110 fs for PTI-LiCa and 64 fs for PTI-LiK. Since Ca²⁺ doping is not expected to alter the band structure substantially, we assume similar effective masses of carriers in both materials, and the longer scattering time in PTI-LiCa thus suggests an enhanced charge carrier mobility, higher intrinsic conductivity, and more delocalized charge nature. Overall, the THz spectroscopy results are fully consistent with the observed high charge carrier mobility in PTI-LiCa, and free carrier-dominated dynamics following light excitations. The transformation of the photogenerated charges carriers from the excitons in PTI-LiK to the free charge carriers in PTI-LiCa was further evidenced by the reversal of the SPV polarity (Supplementary Fig. 29). For PTI-LiK, the excitons with high binding energy would make it rather challenging to form classic depleted space charge layer as well as the upward band bending that exist in typical n-type semiconductors. This was primarily due to the following two reasons. First, the lack of a driving force for the outward migration of charge carriers is evident in PTI-LiK, an n-type polymeric semiconductor. Second, the smaller effective mass of electrons compared to holes enables faster migration of photogenerated electrons than holes in the photocatalysts, leading to a negative signal in PTI-LiK, as shown in Supplementary Fig. 29a. In sharp contrast, the positive SPV signals of PTI-LiCa indicate the formation of upward band bending in its spatial charge layer (SCL), meaning that the charge carrier dynamics in PTI-LiCa closely resemble those of typical metal-contained n-type semiconductors with the photogenerated charge carriers as free charges rather than excitons. That is, in PTI-LiCa, the spontaneous dissociation of excitons to free charges led to the formation of upward band bending in the space charge layer, thus boosting the migration of holes towards the surface under light irradiation, as illustrated by the upper inset in Supplementary Fig. 29a. Such a conclusion is also confirmed by the SPV phase patterns plotted in Supplementary Fig. 29b, in which the signals of PTI-LiCa and PTI-LiK in Quadrant I and Quadrant II correspond to the confirmed preferential migration of holes and electrons, respectively.

The substantial change in charge transport in PTI-LiCa compared with PTI-LiK is also reflected by the TPV spectra in Supplementary Fig. 29c. While the long negative TPV extremum of PTI-LiK at 0.73 ms was the result of the slow diffusion of strongly bound excitons, in sharp contrast, it took only 0.34 µs for PTI-LiCa to reach the positive TPV extremum, three orders of magnitude faster. The positive SPV responses of PTI-LiK within 390 ~ 400 nm and 300 ~ 315 nm are attributed to the surface-state-related depopulation transition (Supplementary Fig. 29d) and the dissociation of high-energy excitons induced by high-energy UV photons. In addition, the enhanced charge separation and transport in PTI-LiCa were also confirmed by its substantially lowered intensity and shortened lifetime of luminescence as revealed in steady-state and transient photoluminescence spectra (Supplementary Fig. 30 and Supplementary Table 4). The fitting results of TR-PL spectra further validate both the high charge separation efficiency in PTI-LiCa and the suppression of defect-mediated recombination via trap-state capture/release processes, which is consistent with the photophysical and structural characterization, as shown in Supplementary Fig. 31. Besides, after depositing cocatalysts (CoOx/Pt@Cr2O3), both samples showed negative SPV signals due to the extraction of photogenerated electrons by the Pt cocatalysts (Supplementary Fig. 32).

To further support our spectroscopic results, we performed femtosecond transient absorption spectroscopy (fs-TAS). As shown in Fig. 3d–g, we present time-resolved contour plots and selected spectra at various delay times following 365 nm excitation. For both PTI-LiCa and PTI-LiK, the negative fs-TAS signals observed between 500–580 nm correspond to stimulated emission (SE), consistent with their photoluminescence spectra, which show emission across the 400–600 nm range. Beyond 580 nm, a broad positive feature appears, which we attribute to excited-state absorption (ESA). In both samples, strong SE and ESA signals emerge within the first 150 fs, indicating rapid generation of electron–hole pairs upon excitation. On the sub-picosecond timescale, the ESA signal in PTI-LiK decays rapidly. Assuming the ESA arises from intra-band free carrier absorption, this fast decay is in good agreement with the sub-ps exciton formation dynamics observed by THz spectroscopy. In contrast, PTI-LiCa exhibits a much longer-lived ESA signal, suggesting that free carriers are generated almost instantaneously after photoexcitation, again in full agreement with the THz results. Taken together, the fs-TAS and THz measurements strongly support our conclusion that photogenerated excitons in PTI-LiCa spontaneously dissociate into free carriers. These findings further underscore the distinct photoinduced charge dynamics in the two materials: PTI-LiCa exhibits free carrier–dominated behavior, while PTI-LiK favors bound exciton formation (Fig. 3h).

For polymeric semiconducting photocatalysts, such as covalent organic frameworks (COFs), graphitic carbon nitride (GCN), poly(heptazine imide) (PHI) and PTI, many attempts have been made to reduce the exciton binding energy, as shown in Fig. 3i8,13,14,39,40,41,42,51,52,53. Although strategies like embedding suitable donor-acceptor pairs have proven effective in reducing the exciton binding energy of the raw COF material from 88 meV to 22 meV, the lowest exciton binding energy in complex carbon-nitride-based polymeric photocatalysts, especially those capable of overall water splitting, such as PTI, stands at 43 meV. Here, the exciton binding energy of 15.4 meV achieved in PTI-LiCa not only reduces the exciton binding energy of this highly promising polymeric semiconducting system to far below the room temperature thermal fluctuation energy (25.7 meV at 298.15 K), thereby generating free charges, but also represents the lowest binding energy of polymer semiconductor photocatalytic materials reported so far. In Supplementary Table 5, we summarize the exciton binding energies of typical metal oxides and polymeric photocatalytic materials reported recently for overall water splitting, highlighting that this work is to demonstrate both spontaneous exciton dissociation and overall water splitting activity.

The underlying reasons for the spontaneous exciton dissociation in PTI-LiCa

For polymeric semiconductors mainly composed of covalent frameworks, both the strong localization of covalent bonds and the lack of a strong dipole moment are the underlying reasons for the high exciton binding energy. For PTI-LiCa, where excitons spontaneously dissociate into free charges, the influences of the lattice contraction and the electrostatic effects of Ca2+ dopants will be explored through theoretical simulations. As shown in Fig. 4a–c, the locations of highest occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) of the unmodified PTI, PTI with intralayer lattice contraction (labeled as PTIcontr.), and the PTI model with both intralayer contraction and Ca dopants (labeled as PTIcontr.:Ca, see Supplementary Fig. 33), were plotted based on theoretical calculations.

Fig. 4: Simulation on the energy states delocalization and anisotropic distribution induced by lattice contraction and Ca doping.
Fig. 4: Simulation on the energy states delocalization and anisotropic distribution induced by lattice contraction and Ca doping.
Full size image

a–c Partial charge density plots of the LUMO and HOMO for pristine PTI, PTI with lattice contraction (PTIcontr.), and PTI with lattice contraction and Ca doping (PTIcontr.:Ca). d Schematic diagram of the exciton dissociation to free charges initiated by the electrostatic field of Ca2+ in the contracted PTI planes.

For the unmodified PTI, it can be observed that both the HOMO and LUMO states were highly localized around the atoms in the triazine rings, with no regionally preferential distribution of significant magnitude, which explained the high exciton binding energy of PTI. As the intralayer lattice contraction was introduced, the densities of HOMO and LUMO states in PTIcontr. increased substantially. Especially, its LUMO states tend to be delocalized and exhibit a preferential distribution in certain regions, due to the increased valence electron wavefunction overlap between different units with shortened lattices. Moreover, both the intralayer contraction and Ca2+ doping introduced in the PTIcontr.:Ca futher extend the delocalization degree of the HOMO and LUMO states with increased densities, indicating substantially enhanced transport of photo-induced charge carriers. Note that the energy state delocalization in PTIcontr. and PTIcontr.:Ca compared with unmodified PTI could also be observed from the side-view states of the HOMO and LUMO (Supplementary Fig. 34).

More importantly, the feature of regional distribution preferences in delocalized HOMO and LUMO states in PTIcontr.:Ca underpins the spontaneous dissociation of excitons into holes and electrons and their subsequent spatial separation. The investigation of the LUMO and HOMO states of the PTI model with Ca2+ doping (labeled as PTI:Ca) (Supplementary Fig. 35) indicates that the Ca doping itself has a marginal impact on the delocalization of energy states, but does bring about a substantial influence on the regionally preferential distribution of LUMO and HOMO states. Thus, it can be concluded that the intralayer lattice contraction is responsible for the delocalization of states across the multiple triazine rings, and the Ca2+ doping with a strong electrostatic field induces the anisotropy, as illustrated in Fig. 4d.

Chemically traceable spatial charge separation in PTI-LiCa

The obvious anisotropic distribution of the HOMO and LUMO states in PTIcontr.:Ca makes the spatial separation of the reduction and oxidation active sites on the surface of PTI-LiCa nanoplates. Here, the spatial separation of photo-induced charge carriers in PTI-LiCa nanoplates was observed through the spatially selective photo-deposition of oxidation and reduction solid products. The high-angle annular dark-field scanning transmission electron microscope images (HADDF-STEM) and the energy-dispersive X-ray spectroscopy (EDS) mapping images of a representative PTI-LiCa nanoplate with photo-deposited with Pt/Cr2O3 (reduced by electrons from Pt4+ and Cr4+) and CoOx (oxidized by holes from Co2+), are shown in Fig. 5a–d and Supplementary Fig. 36. While the signals of Pt and Cr are located evenly on the {10\(\bar{1}\)0} facets of the nanoplate, the Co signals is concentrated preferentially on the lateral edges of the hexagonal nanoplate. This is different with the mixed distribution of Pt and CoOx on the lateral edges of PTI prisms synthesized in LiCl/KCl eutectic salts. Besides, the SEM images of PTI-LiK and PTI-LiCa particles with photo-deposited Au were used to trace the reductive sites, as displayed in Supplementary Figs. 37 and 38. The Au particles were found to mainly deposit on the lateral edges of PTI-LiK prisms, but were evenly deposited on the (\(10\bar{1}0\)) planes of PTI-LiCa nanoplates, consistent with the HADDF-STEM/EDS observation. Particularly, the absence of photo-deposited Au particles on the edges with lower coordination also ruled out the arrival of photo-induced electrons to the edge of PTI-LiCa. Based on the photo-deposition results, it was concluded that the photo-induced electrons and holes in PTI-LiCa tend to migrate towards the {10\(\bar{1}\)0} planes and the lateral edges of the hexagonal nanoplates, respectively, as illustrated in Fig. 5e.

Fig. 5: The spatial charge separation in PTI-LiCa.
Fig. 5: The spatial charge separation in PTI-LiCa.
Full size image

a–d HADDF-STEM images and EDS elemental mapping of a PTI-LiCa nanoplate deposited with Pt/Cr2O3 and CoOx. e Schematic diagram for anisotropic transport of photo-induced electrons and holes, and the spatial distribution of the reduction and oxidation sites on the surfaces of the PTI-LiCa nanoplate. f Photocatalytic OWS activities of PTI-LiCa and PTI-LiK dependent on the deposition sequences of reduction cocatalysts (Pt/Cr2O3) and oxidation cocatalyst (CoOx). Source data are provided as a Source Data file.

As an interesting consequence of the spatial charge separation in PTI-LiCa, it was found that the photo-deposition sequence of the reductive and oxidative cocatalysts no longer affects the OWS performance of PTI-LiCa, as shown in Fig. 5f and Supplementary Fig. 39. However, for the PTI prisms without spatial charge separation, the sample with deposited Pt prior to CoOx exhibited the sharply decreased OWS performance than that of PTI with reversed cocatalysts deposition sequence, as illustrated in the Fig. 5f. This phenomenon occurred because the sites available for oxidation reaction would be firstly occupied and the oxidation reaction was the decisive step in the photocatalytic reaction. As reported previously, if the reductive cocatalyst Pt was first deposited before CoOx, the sites available for oxidation reactions would be occupied, thus leading to a sharp decrease in photocatalytic performance8. In contrast, such a dependence was not observed in the case of PTI-LiCa when the deposition sequence of the two cocatalysts was changed. It is noteworthy that this phenomenon was observed in the PTI system, to our knowledge.

Discussion

The PTI-LiCa hexagonal nanoplates, as a class of promising low-cost carbon-nitride OWS photocatalyst, were synthesized in LiCl/CaCl2 eutectic salts. The (\(\bar{1}\)12) planes of CaCl2 acting as the nucleation and growth substrates induced the lattice contraction and the following delocalization of HOMO and LUMO states within the PTI (001) planes. The Ca dopants simultaneously initiated the charge transfer anisotropy in the contracted carbon-nitrogen frames. Thus, the spontaneous dissociation of excitons to free charges, and the spatial separation of electrons and holes between the (10\(\bar{1}\)0) facets and the lateral side-edges were realized in PTI photocatalysts. Benefited from the effective and fast separation and anisotropic photogenerated charge carriers, the PTI nanoplates with a lower exposure percentage of active (10\(\bar{1}\)0) facets showed enhanced photocatalytic OWS performance, which is about 5 times higher than the counterpart PTI hexagonal prisms with the exposure of (10\(\bar{1}\)0) facets synthesized from LiCl/KCl eutectic salts. The strategy combining the lattice contraction and high-valence cation doping is a promising strategy for developing various polymeric-based semiconductors with large exciton binding energies and no anisotropic transport of charges.

Methods

Materials

Dicyandiamide (DCDA, C2H4N4, 99%), K2CrO4 (≥ 99%) and H2PtCl6 (99.995%) was purchased from Sigma-Aldrich Chemical Reagent Co., Ltd. LiCl (99%) and CoCl2 (99.7%) was purchased from Alfa Aesar Chemical Reagent Co., Ltd. KCl (≥ 99.5%) and CaCl2 (≥ 96%) was purchased from Sinopharm Chemical Reagent Co., Ltd. All reagents were used without further purification.

Sample preparation

Preparation of Poly(triazine imide)(PTI): Typically, DCDA(1 g) was finely mixed with 10 g CaCl2/LiCl eutectic mixture in a glove box under a high-purity argon atmosphere. The mixture was transferred into a quartz tube (Φout = 2.2 cm, where Φ represents the diameter, length = 12 cm) and then heated to 400 oC for 6 h with a ramp rate of 2 oC/min in a muffle furnace. After cooling down to room temperature, the quartz tube was vacuumized and then sealed. Afterward, the sealed tube was thermal heated to 550 oC for 24 h with a ramp rate of 1 oC/min in a muffle furnace. After naturally cooling down to room temperature, the tube was opened carefully and taken out. Then, the products were transferred to a beaker and thoroughly washed with boiling deionized water for at least three times. Finally, the resulting products were collected by drying in the vacuum oven at 60 oC overnight. The sample was simply denoted as PTI-LiCa. For the synthesis of PTI-LiK, the synthesis procedure is similar to that of PTI-LiCa. Other conditions are guaranteed except for the species of the eutectic salts (LiCl/KCl).

Materials characterization

X-ray diffraction patterns of the samples were collected on a Rigaku D / max 2400 spectrometer using Cu kα irradiation. FTIR spectra were collected on a Bruker Tensor 27. The morphology and microstructures of the sample powders were investigated by field-emission scanning electron microscopy (FESEM) (Verios G4 UV, Thermo Scientific), transmission electron microscope (TEM), high-resolution transmission electron microscope (HRTEM) and Energy-dispersive X-ray spectroscopy mapping images (EDS) elemental (FEI Tecnai-F20 and JEM-ARM200F), respectively. The Solid-state 13C NMR measurements were conducted on a Bruker Advance NEO 600 WB spectrometer. Raman spectra were collected by LabSpec 5 Raman spectrometer with 325 nm laser. The composition and chemical states of the samples were analyzed with XPS (Thermo Escalab 250XI with a monochromatic Al Kα X-Ray source). All binding energies were referenced to the C 1 s peak (284.6 eV) arising from adventitious carbon. The Ultraviolet-Visible (UV-Vis) diffuse reflectance spectra were collected on a spectrophotometer (JASCO V-770, Japan) with an integrating sphere. Photoluminescence emission spectra (365 nm excitation) were measured at room temperature (Edinburgh Instruments, FlSP-920). The Ca K-edge XAFS spectra were collected at 4B7A Medium Energy X-ray Beamline of Beijing Synchrotron Radiation Facility (BSRF) and BL16U1 Tender X-ray Spectroscopy Beamline of Shanghai Synchrotron Radiation Facility (SSRF). The in situ FTIR spectra were obtained by using an infrared spectrometer (Thermo Scientific Nicolet iS50). The fs-TAS data was collected on a commercial transient absorption spectrometer (HELIOS, Ultrafast systems). TOF-SIMS data was carried out on ION TOF-SIMS 5. Ultrafast THz spectroscopy was characterized according to a previously reported method54.

A typical three-electrode system, in which Ag/AgCl was selected as the reference electrode, Pt was selected as the counter electrode, and FTO cohered with PTI samples were selected as the working electrode, and was used to measure the Mott-Schottky (MS) plots in a cell containing 0.2 M Na2SO4 aqueous solution (pH = 6.8 ± 0.2) (prepared freshly before use). Typically, 70 mg PTI samples were pressed into a wafer with a diameter of 8 mm and then cohered into an FTO glass using the conductive silver paste.

The surface photovoltage (SPV) spectra measurements were conducted at room temperature (PerfectLight PL-SPV1000) and consisted of monochromatic light, a lock-in amplifier (SR830, Stanford Research Systems) with a light chopper (SR540, Zolix), a photovoltaic cell, and a computer. A 500 W Xe lamp serves as the light source, and the monochromatic light is provided by passing light from the Xe lamp through a grating monochromator. A low chopping frequency of 24 Hz was used in the testing process. The structure of the SPV sample cell is an ordered mica-ITO-sample-copper platform.

The Transient photovoltage (TPV) spectra were measured at room temperature. This system included a Nd:YAG laser source(Q-Smart 450, Quantel) emitting at 355 nm, a 500 MHz digital phosphor oscilloscope (TDS 5054, Tektrionix) equipped with a preamplifier (5003 Brookdeal Electronics), and a parallel-plate capacitor-like sample chamber. The Flash Lamp Q-Switch delay of the laser was adjusted to 100 μs, repetition rate of 5 Hz and at a wavelength of 355 nm for this investigation.

Photocatalytic activity characterization

Photocatalytic water splitting was conducted in an automatic testing system (Perfectlight Sci&Tech, Labsolar-6A) with a 250 ml reaction container and a 300 W Xe lamp as a light source. The temperature of the reaction solution was maintained at 10 oC by the flow of cooling water during the reaction. Notably, three sequential photo-deposition steps were performed before testing the photocatalytic overall water-splitting, and all measurements were conducted under full-spectrum irradiation. 100 mg of the photocatalysts were dispersed in 100 ml deionized water. Firstly, the 0.5 wt% CoOx as an oxygen evolution cocatalyst was achieved by dissolving CoCl2(1 mg/ml) in the above 100 ml reaction solution, and the reactor was evacuated and irradiated for 1 h. Then the reactor was opened, and the 1 wt% Pt as a reducing cocatalyst was in situ photodeposited by dissolving H2PtCl6 (1 mg/ml) in the above solution and irradiated with 1 h. The 1 wt% Cr2O3 was acquired as same as above except using K2CrO4 (1 mg/ml) as a precursor. Afterward, the reactor was evacuated several times to remove air completely prior to irradiation under a 300 W xenon lamp (full spectrum).

Apparent quantum efficiency (AQE) measurement

The 300 W Xe Lamp with an irradiation area of π × (3.5 cm)2 was controlled as a light source to estimate the AQE. The photo flux used in the AQE equation was determined by the weighted average of photo flux at five sample points. Note that the weight value of the center sampling point was 2/3, and the average weight values of the other four sampling points on the circumference were 1/3. The amount of hydrogen produced by photocatalytic overall water splitting was measured at a monochromatic wavelength.

$${\rm{AQE}}=\frac{{\rm{The}}\; {\rm{amount}}\; {\rm{of}}\; {\rm{Hydrogen}}\times 2\,}{{\rm{Incident}}\; {\rm{photon}}\; {\rm{flux}}}$$
(1)

Electrocatalytic activity characterization

The electrochemical measurements were conducted on a three-electrode system and controlled by a CHI 760 electrochemical station. A GC electrode connected to a Pine Modulated Speed Rotator was used as the working electrode. The catalysts were uniformly loaded on the conductive carbon disk of 0.196 cm2 by 0.2 mg cm−2. A graphite rod and an Hg/HgO electrode were used as the counter and reference electrode, respectively. The electrolyte used for HER, ORR, OER was Ar-, O2-, and O2-purged 1.0 M KOH solution (pH = 13.8 ± 0.1), respectively. The linear sweep voltammetry (LSV) was conducted with a scan rate of 5 mV/s, and all potentials were referenced to the reversible hydrogen electrode (RHE), following the below equation:

$${E}_{{\rm{RHE}}}={E}_{{\rm{Hg}}/{\rm{HgO}}}+0.098\,{\rm{V}}+0.059\,{\rm{V}}\times {\rm{pH}}$$
(2)

Theoretical calculations

For the theoretical simulation of the location of Ca2+ dopants in the framework of PTI and HOMO and LUMO states, the DFT calculations were performed using the linear-scaling self-consistent field algorithm55 within the CP2K/Quickstep framework56. This method integrates computational efficiency with high precision, advancing large-scale quantum simulations. In this study, a hybrid Gaussian and plane wave scheme was used for the basis set, with BASIS-SET designated as DZVP MOLOPT-SR-GTH. In addition, all simulations were executed using a 1 × 1 × 1 Gamma-centered Monkhorst-Pack electronic k-point mesh, with a plane-wave cut-off energy of 500 eV. Moreover, the surface model incorporates an approximate 15 Å vacuum region along the z-axis, and employs the DFT-D3 dispersion correction method, as proposed by Grimme, to account for van der Waals interactions57. The atomic coordinates of the optimized crystal structures can be found in Supplementary Data 1. Finally, the convergence criteria for energy and force per atom were set to 10−6 eV and 0.03 eV/Å, respectively.

As for the theoretical explanation for morphology transformation from hexagonal prisms to nanoplates due to the utilization of CaCl2 for PTI-LiCa, the DFT calculations were performed using the Vienna ab-initio Simulation Package (VASP)58. The generalized gradient approximation (GGA) combined with the Perdew-Burke-Ernzerhof (PBE) functional was adopted to describe the exchange correlation interaction59. The electron-ion interactions were described by the projector augmented wave (PAW) method60. All geometries were fully relaxed with the convergence criterion of 10−6 eV and 0.01 eV·Å−1 for the total energy and residual force, respectively.