Photon upconversion through triplet exciton-mediated energy relay

Exploration of upconversion luminescence from lanthanide emitters through energy migration has profound implications for fundamental research and technology development. However, energy migration-mediated upconversion requires stringent experimental conditions, such as high power excitation and special migratory ions in the host lattice, imposing selection constraints on lanthanide emitters. Here we demonstrate photon upconversion of diverse lanthanide emitters by harnessing triplet exciton-mediated energy relay. Compared with gadolinium-based systems, this energy relay is less dependent on excitation power and enhances the emission intensity of Tb3+ by 158-fold. Mechanistic investigations reveal that emission enhancement is attributable to strong coupling between lanthanides and surface molecules, which enables fast triplet generation (<100 ps) and subsequent near-unity triplet transfer efficiency from surface ligands to lanthanides. Moreover, the energy relay approach supports long-distance energy transfer and allows upconversion modulation in microstructures. These findings enhance fundamental understanding of energy transfer at molecule-nanoparticle interfaces and open exciting avenues for developing hybrid, high-performance optical materials.

Here we demonstrate that it is possible to achieve enhanced upconversion emission through triplet exciton-mediated energy relay (Fig. 1a). Excitation energy is harvested and upconverted via energy transfer in the nanoparticle core. Subsequently, emission in the ultraviolet spectral region is absorbed by capping ligands, followed by back-energy transfer to activators in the shell. This process gives rise to the observed upconversion emission (Fig. 1b). In our design, the surface ligand bridges excitation energy transfer from the core to the shell layer. A suitable organic ligand must satisfy the following criteria: strong broadband absorption in the ultraviolet; a good match between its lowerlying excited states and the activator's emitting states for efficient back-energy transfer; robust surface coordination and high photostability.
Upconversion emission strongly depends on the number of ligands and lanthanide activators. Here we take CPPOA as an example. In principle, an increased number of CPPOA ligands improves excitation energy harvesting from Yb-Tm pairs in nanoparticles. However, a high loading density of ligands often suffers from concentration quenching, thus dissipating the excitation energy nonradiatively. Similarly, an increase in the activator number generally favors trapping of excitation energy. However, an elevated number of activators can result in deleterious cross-relaxation 47 . By carefully tuning the number of activators and CPPOA molecules, we obtained a maximum emission intensity with 1-μmol Tb 3+ and 0.5-μmol CPPOA for 15-mg NaYF 4 @NaYbF 4 :Tm@NaYF 4 multilayer nanoparticles (Fig. 2a, b and Supplementary Figs. 9 and 10; Supplementary Table 4). With optimized concentrations, the Tb 3+ emission intensity of CPPOA-engineered nanoparticles enhanced 158 times compared with that of a typical Gd 3+ -mediated upconversion nanosystem under a power density of 108 W/cm 2 . The energy-relay nanosystems showed a low power dependence (Supplementary Figs. 11 and 12). When excitation power was further decreased to 76 W/cm 2 , the energy relay-mediated upconversion system showed an intense Tb 3+ emission, while the energy migration-mediated upconversion system exhibited negligible emission (Fig. 2c). Given that Gd 3+ -based energy migration suffers from luminescence quenching caused by the vibration of surrounding solvent molecules, we tested the optical stability of CPPOA-modified multilayer nanoparticles in various solvents. Surprisingly, these CPPOA-coated nanoparticles showed invariant emission intensity ( Supplementary Fig. 13). Moreover, these nanoparticles exhibited high photostability under 980-nm irradiation ( Supplementary Fig. 14).
Efficient energy transfer between nanoparticles and surfacebound CPPOA molecules is essential for achieving intense upconversion emission. The core-shell-shell design favors efficient energy transfer from Tm 3+ donors confined in nanoparticles to surface-bound CPPOA molecules. In this hybrid nanosystem, each Tm 3+ dopant in the nanoparticle is supposed to be an individual energy donor for transferring excitation energy to all CPPOA acceptors on the particle surface. Thus, overall energy transfer efficiency is determined by averaging all possible distributions of donor-acceptor pairs [48][49][50][51] . In the coreshell-shell design, all energy donors (Tm 3+ ) are spatially confined in the middle shell. Compared to a random distribution of donors in the core-dominated core-shell structure, this coreshell-shell design may offer higher energy transfer efficiency because of a shorter donor-acceptor distance, as revealed by the Monte-Carlo simulation (Fig. 2d, e). The energy transfer from Tm 3+ to CPPOA molecules is simulated based on Förster resonance energy transfer due to large separation distances (> 2 nm) between Tm 3+ and the molecules [52][53][54][55][56] . To validate our hypothesis, we prepared CPPOA-modified NaYbF 4 :Yb/Tm (1 mol%)@NaYF 4 :Tb core-shell nanoparticles as a control. We observed a much higher Tb 3+ emission intensity from core-shellshell nanoparticles than from conventional core-shell nanoparticles ( Fig. 2f). As an added benefit, the core-shell-shell design enables excitation energy to be confined in the NaYbF 4 middle shell, which reduces energy loss to crystal lattice defects and increases local excitation energy density. High-density excitation facilitates the accumulation of excitation energy in high-lying states of Tm 3+ 44 .
Cation-exchange reactions allow a large portion of lanthanide emitters to be exposed at nanoparticle surfaces, enabling direct bonding between surface emitters and capping molecules. Unlike the Tm 3+ -to-CPPOA energy transfer described above, we reasoned that strong exchange coupling dominates energy transfer between molecules and surface lanthanide emitters (e.g., Tb 3+ ) because their proximity results in fast triplet generation of molecules and efficient triplet energy transfer from molecules to lanthanide emitters 57 . To validate this hypothesis, we next performed ultrafast transient absorption spectroscopy to probe the photodynamics of excitation energy at the interface of CPPOA and Tb 3+ -exchanged multilayer nanoparticles. The broadband (900−1300 nm) photo-induced absorption feature of molecular triplets appears on a picosecond timescale (Fig. 3a). Kinetics of spectral components reveals a triplet state rise time of 93 ps ( Fig. 3b and Supplementary Fig. 15), 198 times faster than that obtained from pristine CPPOA (18.4 ns) 57 . Decay of the T 1 state has a constant 10.6 ns ( Fig. 3c and Supplementary Fig. 16), 2547-fold faster than that of CPPOA molecules. The quantum efficiency of triplet transfer from CPPOA ligands to nanoparticles was close to 100% (Fig. 3d). To further verify the cation-exchange effect on energy upconversion efficiency, we performed a control experiment by directly doping Tb 3+ ions into the outmost NaYF 4 shell of the multilayer nanoparticle. These directly doped nanoparticles showed significantly suppressed Tb 3+ emission compared to Tb 3+ -exchanged multilayer nanoparticles (Fig. 3e).
Long-distance energy relay. The large absorption cross-section of the surface ligand provides an opportunity to extract excitation energy from the core over a long distance, achieving multicolor upconversion tuning at a microscopic scale. To test the feasibility of our upconversion process in a microscopic region, we synthesized a set of Tb 3+ -exchanged NaYF 4 :Yb/Tm@NaGdF 4 hedgehog-like microrods. After coating these microrods with CPPOA molecules, we observed cyan emission under 980-nm excitation (Fig. 4a, top panel), characteristic of Tb 3+ emission peaks (Fig. 4a, bottom panel). Similarly, we also achieved Tb 3+ emission from CPPOA-capped NaYF 4 :Yb/Tm@NaGdF 4 :Tb hedgehog-like microplates of 5 μm in diameter (Fig. 4b), which was unattainable by conventional Gd 3+ -mediated energy migration 33 . The long-distance energy relay mediated by surface ligands can also be realized across two sets of nanoparticles. To validate this, we prepared NaYF 4 @NaYbF 4 :Tm@NaYF 4 nanoparticles as energy donors and NaGdF 4 :Tb/CPPOA-tagged polystyrene microspheres as energy acceptors. Donor particles were first deposited on a glass substrate, and acceptor particles were then drop-casted on the donor particle layer. Upon 980-nm excitation, we observed bright green emission from NaYF 4 :Tb/ CPPOA-tagged polystyrene microspheres, suggesting the occurrence of an efficient energy relay process ( Fig. 4c-g).

Discussion
In conclusion, we have demonstrated that photon upconversion from lanthanide activators without ladder-like energy levels can be achieved by a triplet exciton-mediated energy relay. A multilayer core-shell design coupled with specially designed molecules enables precise control over energy transfer and efficient energy relay at the nanoparticle-molecule interface. Compared with Gd 3+ -based energy migration systems, energy relay upconversion offers advantages such as a broad selection of host lattices, high emission intensity, less dependence on excitation power, and high stability in various solvents. Moreover, longdistance energy relay allows upconversion emission to be achieved in microstructures. This development will expand the applicability of lanthanide nanoparticle-molecule hybrid systems for diverse applications.
Synthesis of NaYF 4 core nanoparticles. In a typical procedure, an aqueous solution of Y(CH 3 CO 2 ) 3 •xH 2 O (2.0 mL, 0.2 M) was mixed with 3.0 mL of oleic acid (OA) in a 50-mL flask. The resulting mixture was heated to 150°C in an oil bath and maintained at this temperature for 30 min, at which time 7.0 mL of 1octadecene (ODE) were added to the flask. After the mixture was cooled to 50°C over 30 min, a 6.0-mL methanol solution containing NH 4 F (1.6 mmol) and NaOH (1.0 mmol) was added to the core precursor and stirred for 30 min. After removing volatile solvents, the temperature was increased to 290°C under an argon atmosphere. After heating for 2 h, the mixture was cooled to room temperature and washed several times with ethanol. Products were re-dispersed in 4.0 mL of cyclohexane.
Synthesis of NaYF 4 @NaYbF 4 :1%Tm nanoparticles. The NaYbF 4 :1%Tm shell precursor was prepared by mixing an aqueous solution of Ln(CH 3 CO 2 ) 3 •xH 2 O (2.0 mL, 0.2 M, Ln = Yb or Tm) with 3.0 mL of OA and 7.0 mL of ODE. The mixture was heated at 150°C in an oil bath for 1 h. After cooling to 80°C, NaYF 4 core nanoparticles in 4.0 mL of cyclohexane were added to the mixture and kept at 80°C for 30 min. Subsequently, a methanol solution of NH 4 F (1.6 mmol) and NaOH (1.0 mmol) was added at 50°C and stirred for 30 min. After removing volatile solvents, the mixture was heated to 290°C under an argon atmosphere and maintained at this temperature for 3 h. After cooling to room temperature, the products were precipitated, washed several times with ethanol, and re-dispersed in 4.0 mL of cyclohexane for further use.
Synthesis of NaYF 4 @NaYbF 4 :1%Tm@NaYF 4 nanoparticles. The NaYF 4 shell precursor was prepared by mixing 1.0 mL of Y(CH 3      with 3.0 mL of OA and 7.0 mL of ODE. The mixture was heated at 150°C in an oil bath for 1 h. After cooling to 80°C, the as-prepared NaYF 4 @NaYbF 4 :1%Tm nanoparticles were added to the mixture and kept at 80°C for 30 min. Subsequently, a methanol solution of NH 4 F (0.8 mmol) and NaOH (0.5 mmol) was added under intense agitation at 50°C. After 30 min, volatile solvents were evaporated by heating at 100°C for 15 min. The mixture was then heated at 290°C under an argon atmosphere for 2 h. After cooling to room temperature, nanoparticle products were collected, washed several times with ethanol, and redispersed in 4.0 mL of cyclohexane for further use 44 .
Synthesis of NaYF 4 :Yb/Tm (30/0.5 mol%) microrods. The microrod synthesis was performed according to a hydrothermal method 58 . A mixture of YCl 3 (1.0 mL, 0.2 M), OA (5.0 mL), ethanol (5.0 mL) and NaOH (0.3 g) was stirred at room temperature. After 20 min, an aqueous solution of NH 4 F (2.0 M, 1.0 mL) was added and kept for another 30 min. The mixture was then transferred to a Teflon-lined autoclave (20 mL) and heated at 220°C for 12 h. The precipitate was collected, washed several times with ethanol, and re-dispersed in cyclohexane (4.0 mL) for further use.
Synthesis of 5-μm NaYbF 4 :Tm (1 mol%) microplates. In a typical procedure 58 , a mixture of sodium citrate (0.5 mL; 0.3 M) and YbCl 3 (2.0 mL; 0.2 M) was stirred at room temperature to form a milky suspension. After that, an aqueous solution of NaF (9.6 mL; 0.5 M) was added. The resulting mixture was then transferred to a Teflon-lined autoclave (20 mL) and heated at 220°C for 12 h.
Synthesis of gadolinium trifluoroacetate precursors. In a typical experiment, Gd 2 O 3 (5.0 mmol) was first dissolved in trifluoroacetic acid (10 mL) at 80°C, and then dried at 100°C. The as-prepared lanthanide trifluoroacetate powder was dissolved in water to form gadolinium precursors (0.2 M, 50 mL).
Synthesis of hierarchical microstructures. This synthesis was performed by an epitaxy growth method 59 . Typically, a cyclohexane dispersion of the as-prepared microrod or microplate seeds (1.0 mL) was first added to a mixture of OA (3.0 mL), ODE (3.0 mL), gadolinium trifluoroacetates (0.2 mmol), and sodium trifluoroacetate (0.36 mmol). After heating to 100°C and degassing for 30 min, the mixture was heated under a nitrogen atmosphere to 310°C with a 12°C/min rate and maintained at that temperature for 30 min. The products were collected, washed several times with ethanol, and re-dispersed in cyclohexane for further use. The same protocol was adopted to synthesize hierarchical NaYbF 4 :Tm (1 mol%) @NaGdF 4 microplates except for using the microplates as seeds.
Preparation of ligand-free nanoparticles. The as-prepared hydrophobic nanoparticles were precipitated by the addition of ethanol and then dispersed in an acidic ethanol solution (0.1 M HCl) by ultrasonication. Nanocrystals were collected by centrifugation at 33,481 × g for 20 min. The resulting products were washed several times with ethanol and deionized water, and re-dispersed in deionized water.
Preparation of Tb 3+ -exchanged core-shell-shell nanoparticles. These nanoparticles were prepared by cation-exchange reactions 46 . In a typical synthetic process, the as-prepared ligand-free nanoparticles (500 μL) were mixed with different amounts of TbCl 3 at room temperature. After 24 h, the nanoparticles were collected by centrifugation at 39,846 × g for 30 min and washed several times with ethanol. Finally, the products were re-dispersed in 1.0 mL of ethanol.
Ligand modification to nanoparticle surfaces. An ethanol solution of ligand-free nanoparticles (1.0 mL) was separated into two equal portions. CPPOA ligands were added to the particle solution and ultrasonicated for 1 h to ensure ligand coordination to nanoparticle surfaces. After that, excess ligands were removed by centrifugation. Finally, the products were re-dispersed in ethanol for optical measurement.
Preparation of nanoparticle-tagged polystyrene beads. In a typical experiment 33 , an aqueous solution containing ligand-free nanocrystals (50-100 μL, 2 wt%) was mixed with polystyrene bead dispersion (5.0 μL) and butanol (200 μL) at room temperature. After 30 min, the final products were collected by centrifugation at 4427 g for 5 min, washed with H 2 O, and re-dispersed in deionized water.

Synthesis of 2-(diphenylphosphoryl)-5-bromotoluene (TPPOMBr) molecules.
The synthetic procedure is shown in Supplementary Fig. 17. In a typical experiment, 2.97 g (10.0 mmol) of 2-iodo-5-bromotoluene were dissolved in 20.0 mL of anhydrous ether. The solution was added dropwise to a mixture comprising magnesium chips (264 mg; 11.0 mmol) and a small amount of iodine. The mixture was heated at 40°C for 1 h, and then cooled to 0°C in an ice bath. A diluted solution of diphenylphosphine chloride (2.65 g; 12.0 mmol) in 10.0 mL of anhydrous ether was added dropwise. After addition, the reaction temperature was increased to room temperature and then stirred overnight. The reaction was quenched by adding 20.0 mL of water and extracted by dichloromethane (3 × 30 mL). The organic layers were dried with anhydride Na 2 SO 4 . The solvent was removed in vacuo. The residue was dissolved using 50 mL of dichloromethane. After the addition of 3-mL hydrogen peroxide (30%), the reaction mixture was stirred for 4 h and extracted using sodium hydrogensulfite-saturated solution. The organic layers were dried with anhydride Na 2 SO 4 . The solvent was removed in vacuo. The product was purified by flash column chromatography to afford a white powder in 80% yield 1   Synthesis of 2,2′:6′,2″-terpyridine-6,6″-dicarboxylic acid (TDA). The synthetic procedure for TDA molecules is shown in Supplementary Fig. 18. In a typical procedure 61 , 0.47 g of 2,2′,2″-terpyridine (2.0 mmol), and 20 mL of CH 2 Cl 2 were added to a solution of 1.30 g of m-chloroperbenzoic acid (7.5 mmol) in 20 mL of CH 2 Cl 2 . The mixture was stirred at room temperature for 13 h. After washing with 5% Na 2 CO 3 solution and drying over anhydrous MgSO 4 , the solvent was evaporated and purified by washing with acetone to give the product (0.44 g, 83% yield). A sample product (2.17 g) was mixed with CH 2 Cl 2 (80 mL). Then, Me 3 SiCN (12.9 g; 82 mmol) was added to the mixture. After stirring for five min, PhCOCl (3.66 g; 32.8 mmol) was added over 15 min and the mixture was stirred overnight at room temperature. The mixture was neutralized with 10% K 2 CO 3 solution, and the aqueous phase was extracted with CHCl 3 . The combined organic phase was evaporated and the residue was crystallized from MeCN/THF for further use. Subsequently, the resulting compound (300 mg) was added to a solution containing ethanol (19.0 mL), water (4.0 mL) and KOH (594 mg). The reaction mixture was refluxed overnight, and then the solvent was evaporated under vacuum. The residue was dissolved in water (5 mL) and the pH was adjusted to 4 with aqueous HCl (1 M). The precipitate was removed by filtration and washed with cold water and acetonitrile. After that, the solid material was heated to reflux in a mixture of H 2 SO 4 /CH 3 COOH (9 mL, 1:1) for 5 h and the reaction mixture was poured onto ice (45 mL). The product was filtered and washed with cold water and acetonitrile to yield the product (TDA, 312 mg, 92% yield) as a white solid 1  Monte-Carlo modeling of energy transfer from Tm 3+ to surface ligands. The energy transfer process is described based on a Förster resonance energy transfer (FRET) mechanism. Monte-Carlo calculations of energy transfer efficiency were executed using Mathematica 12 and performed in a simulated 3D coordinate matrix. In CPPOA-modified nanoparticles, each Tm 3+ ion can be simplified as a point randomly distributed within a nanoparticle and serve as an individual energy donor. Each CPPOA ligand can be assumed as a point-like energy acceptor randomly distributed on the particle surface. The Monte-Carlo simulation was generated using the following criteria. First, 5000 Tm 3+ ions are randomly distributed in the core of a core-shell structure or confined within the middle shell of a coreshell-shell structure, while 2000 dye molecules are randomly distributed on particle surfaces. Therefore, every Tm 3+ ion can transfer the excited energy to all surfacebound dyes. As such, each Tm 3+ -CPPOA pair has a specific separation. The calculated FRET efficiency is the assumption of efficiencies for 2000 energy transfer events, which is defined as the probability of finding Tm 3+ ions that can transfer the excitation energy to all surface-bound dye molecules. Using the same method, we can derive all the probability of finding 5000 Tm 3+ donors to offer FRET in a core-shell or core-shell-shell particle. Thus, we can plot the probability of finding energy donors against the spatial distance from the particle surface.
Quantum mechanical calculations. Density functional theory (DFT) calculations were performed for geometry optimization and harmonic frequency analysis of CPPOA, TPDA and TDA molecules using Gaussian 16 (A.03) program 62 . We chose M06-2X functional in combination with 6-311G** basis set 63,64 . No imaginary frequency was detected for these optimized molecular structures. Timedependent density functional theory (TD-DFT) calculations were performed to obtain S 1 and T 1 energy states of the three molecules. For excited-state calculations, we employed the same functional and basis set as ground-state calculations.
Characterizations. The structures of as-synthesized molecules were confirmed by nuclear magnetic resonance (NMR) and high-resolution mass spectroscopy 1 H-NMR, 13 C-NMR, 31  Powder X-ray diffraction (XRD) data were measured with an X-ray diffractometer (Bruker D8 Advance) with graphite monochromic CuKα radiation (λ = 1.5406 Å). Low-resolution transmission electron microscopy (TEM) was performed on a transmission electron microscope (JEOL-1400) operating at an acceleration voltage of 120 kV. High-resolution TEM images were taken using a transmission electron microscope (JEOL-JEM 2010) operated at an acceleration voltage of 200 kV. Energy-dispersive X-ray (EDX) spectroscopic analysis was performed using an energy system (Oxford INCA) operated at 200 kV. UV-Vis spectra were recorded on a UV spectrophotometer (Shimadzu 2450). Fourier transform infrared (FT-IR) spectroscopy employed an FT-IR spectrometer (Varian 3100). Photoluminescence spectra were recorded in a fluorescence spectrophotometer (Edinburgh FSP920) equipped with a photomultiplier (PMT) in conjunction with a 980-nm diode laser and a xenon arc lamp (Xe900). The lanthanide concentration was determined by a Perkin Elmer Avio 500 inductively coupled plasma optical emission spectrometer (ICP-OES).
Transient absorption spectroscopy. Picosecond time measurements were performed using a commercial transient absorption spectrometer (HELIOS, Ultra-fastSystems). Pump pulses (355 nm) were generated through a TOPAS-Prime amplifier (Light Conversion), which was pumped by a 790-nm pulse laser from a regenerative Ti:Sapphire amplifier system (Spectra Physics, Solstice). The broadband probe pulse (800-1550 nm) was obtained by introducing a 790-nm laser pulse into a YAG crystal. The pump-probe delay was computer-controlled with a piezoelectric translation stage. The time resolution of laser pulses was approximately 200 fs. Nanosecond time measurements were performed with an electronically controlled delay.
Samples were excited with a pump pulse and probed at different delay times using a broadband probe pulse. A differential transmission (ΔT/T) signal in transient absorption spectra was recorded over short pulses (500 fs to 6 ns) with a probe covering (500-850 nm, 750-1600 nm) and long (1 ns to 1 ms) time delays with a probe pulse covering (350-750 nm, 850-1020 nm). To identify different components from transient absorption data, a genetic algorithm analysis was also used to distinguish different spectral species and corresponding kinetics. Triplet energy transfer efficiency can be calculated by η = (τ donor only − τ donor-acceptor) / τ donor only . Note that τ donor only refers to the triplet lifetime of the donor in the absence of acceptors, while τ donor-acceptor represents the triplet lifetime of the donor in the presence of acceptors.

Data availability
All data supporting the findings of this study are publicly available from the University of Cambridge repository at https://doi.org/10.17863/CAM.69276.