Transferring the entatic-state principle to copper photochemistry

Abstract

The entatic state denotes a distorted coordination geometry of a complex from its typical arrangement that generates an improvement to its function. The entatic-state principle has been observed to apply to copper electron-transfer proteins and it results in a lowering of the reorganization energy of the electron-transfer process. It is thus crucial for a multitude of biochemical processes, but its importance to photoactive complexes is unexplored. Here we study a copper complex—with a specifically designed constraining ligand geometry—that exhibits metal-to-ligand charge-transfer state lifetimes that are very short. The guanidine–quinoline ligand used here acts on the bis(chelated) copper(I) centre, allowing only small structural changes after photoexcitation that result in very fast structural dynamics. The data were collected using a multimethod approach that featured time-resolved ultraviolet–visible, infrared and X-ray absorption and optical emission spectroscopy. Through supporting density functional calculations, we deliver a detailed picture of the structural dynamics in the picosecond-to-nanosecond time range.

Main

The entatic-state principle has been applied for 50 years to interpret thermally activated electron-transfer processes at copper centres^{1,2,3,4,5,6,7,8,9,10}. Entasis denotes a structural pre-distortion of a transition metal complex towards a reaction transition state, and thus facilitates a chemical reaction and, in a narrower sense, enables a faster electron transfer (Fig. 1a). This pre-distortion is also discussed as the energization of reactive states and is crucial for an efficient catalysis in chemistry and biology^1,8,9. Parallel to this development, photochemistry focuses on the elucidation of excited-state dynamics and reaction pathways^11,12,13. Both concepts have not yet been combined, although the entatic-state principle and photochemical research have found vital applications, especially in copper chemistry because copper is one of the most-important redox-active metals given its central role in several biological processes^14,15. Also, not only is the tuning of the Cu(II/I) redox potential crucial for an efficient electron transfer in nature, it is also important for synthetic complexes in catalytic processes and in solar energy conversion^11,16.

Figure 1: Model complexes for the entatic state and their spectroscopic features.

a, Simplified illustration of the entatic-state principle: Cu(II) has a clear electronic preference for square-planar coordination, whereas Cu(I) does not have an electronically favoured coordination mode (d¹⁰ configuration), but is sterically mostly driven to tetrahedral motifs. The entatic-state principle denotes an energization of both states (here the misfit of ligands to square-planar and tetrahedral coordination), which leads to a smaller activation barrier^9,17. The principle was developed for copper type-1 proteins, but applies also to small-molecule complexes with suitable ligands⁹. b, Overlay of the experimental structures of the cations of 1 and 2 (turquoise, copper; blue, nitrogen; black, carbon). The difference between the Cu(I) and Cu(II) oxidation states is expressed in a change of the angle between the CuN₂ planes by 20° and a shortening of the Cu–N_gua bond length by 0.1 Å (ref. 10). The displayed formula of 1 is also shown. c,d, Static UV/vis spectra of 1 (black) and 2 (red) in MeCN (c) and static infrared spectra (1 (black) and 2 (red)) in CH₂Cl₂ (d). gua, guanidine-related bands; qu, quinoline-related bands; asym, asymmetric; sym, symmetric.

The entatic-state principle uses a pre-organization of the ligand sphere by a constrained geometry with a high similarity for both oxidation states Cu(I) and Cu(II) (ref. 17). Ligand pre-organization is the tendency of a ligand to stabilize a certain coordination mode, in some cases for a specific metal ionic state. This is different to metal–ligand complementarity in which the metal–ligand pair has an ideal geometrical and electronic fit. Comba and Schiek give a detailed discussion of the relation between the entatic state, ligand pre-organization and complementarity¹⁷. The entatic state lowers the reorganization energy during redox processes and, thus, facilitates them (Fig. 1a)^9,18,19. For solar energy conversion schemes it is, however, actually desirable to generate long-lived charge-separated states with sterically demanding ligand structures to enhance the current flow^11,20. Therefore, the entatic-state principle concerns both directions (to enhance and to hamper charge transfer), because excited-state lifetimes are susceptible towards substitution at specific ligand positions and can be tuned over two orders of magnitude by substitution exchange in acetonitrile^21,22.

In the context of the entatic state, intensive studies on conformationally invariant Cu(II/I) model complexes have been performed^{23,24,25,26,27}. Very recently, Policar and co-workers developed a sugar-based ligand that provides a strong pre-organization for Cu(II) coordination in a trigonal bipyramid, but with an unusual stabilization of its Cu(I) redox state²⁸. Dahl and Szymczak synthesized a constrained Cu(I) complex very close to a square-planar coordination²⁹. Recently, we reported a series of bis(chelate) Cu(I) and Cu(II) guanidine–quinoline complex cations [Cu^II(TMGqu)₂]⁺ and [Cu^I(TMGqu)₂]²⁺ (TMGqu, 1,1,3,3-tetramethyl-2-(quinolin-8-yl)-guanidine) (Fig. 1b shows the overlay of both cations¹⁰). They display the astonishing feature that their structures are very similar with a coordination polyhedron in the middle, between tetrahedral and square-planar environments¹⁰. A resonance Raman study of these Cu(II/I) complexes in solution showed that they come into resonance at nearly the same energy, around ~3.5 eV, by metal-to-ligand charge-transfer (MLCT) and ligand-to-metal charge-transfer (LMCT) processes¹⁰. We found a dominant Cu–N vibrational mode that couples the optical charge-transfer excitation with the distortion along the reaction coordinate that leads from the more tetrahedral Cu(I) to a more flattened (towards square-planar coordination) Cu(II) geometry. The donor interplay between guanidine and quinoline units as well as the large steric encumbrance of guanidines were found to be crucial for this constrained coordination. These model compounds are extremely susceptible to MLCT and LMCT processes, as the pre-distortion lowers the energy barrier required to enable charge transfer.

In this work, we report on the dynamics of the structural and electronic changes as induced by an MLCT photoexcitation process by utilizing a collection of complementary experimental transient techniques, which provides crucial information on timescales that cover more than four orders of magnitude. With time-resolved optical absorption and emission spectroscopy in the visible and ultraviolet range we identify short-lived electronic intermediate states. Time-resolved infrared spectroscopy^30,31,32,33 characterizes these intermediates by probing the molecular vibrations in the ligand system. Finally, transient pump–probe X-ray absorption spectroscopy (XAS)^{34,35,36,37,38,39,40,41} focuses on the changes of the copper oxidation state and its coordination sphere in [Cu^I(TMGqu)₂]⁺ after photoexcitation. The combination of these different experimental tools with extensive density functional theory (DFT) studies of the excited states and their spectroscopic features leads to a new comprehensive picture of the reaction dynamics that involves excited singlet and—by intersystem crossing (ISC)—triplet states.

Results and discussion

As entatic-state models, the guanidine–quinoline complexes [Cu^I(TMGqu)₂]PF₆ (1) (Fig. 1b) and [Cu^II(TMGqu)₂](OTf)₂ (2) consist of highly similar complex cations, although they contain copper ions in different oxidation states (structural data are reproduced in Supplementary Section 2 (ref. 10)). The ultraviolet–visible (UV/vis) spectra are dominated by the LMCT transition (quinoline-π* → Cu) of 2 at 390 nm versus the MLCT transition of 1 (Cu → quinoline/guanidine-π*) near 450 nm (Fig. 1c)^10,42. The static infrared spectra (Fig. 1d) of both complexes depict the key vibrations within the guanidine and quinoline units, which are highly sensitive towards the copper oxidation state. These differences are used below to identify short-lived intermediate states. Hereby, the most important bands are attributable to stretching vibrations of the guanidines, which are sensitive to the oxidation state of the copper centre. To monitor the MLCT dynamic processes in 1, we used (1) transient UV/vis and infrared spectroscopy, which allows insights into the fast reaction dynamics directly after excitation up to 500 ps, and (2) time-resolved optical emission spectroscopy and (3) XAS to reach out to the longer nanosecond timescales. In addition, XAS delivers unique structural information on selected excited states, which are directly correlated to density functional studies. All experimental and theoretical details are given in the Supplementary Information.

Time-resolved optical absorption spectroscopy in the visible and infrared regions

Information on the fast reaction dynamics is obtained from time-resolved optical absorption experiments, in which 1 is excited with short pulses (90 fs) at λ_exc = 400 nm in its MLCT region. Figure 2a summarizes the absorption difference spectra recorded in the UV/vis range for a series of delay times (width of the instrumental response function, 150 fs). Directly after excitation, the ground-state absorption in the Cu(I) band decreases at about 450 nm. A very broad absorption increase can be observed at longer wavelengths (480–730 nm) which we assign, supported by DFT calculations, to the formation of a quinoline radical anion upon MLCT⁴³. At the small excitation energy of 400 nJ used in the experiment, solvated electrons do not show up (Supplementary Section 3). The dip in the 530 nm range could originate from a stimulated emission from the excited states (Fig. 2c).

Figure 2: Time-resolved UV/vis and infrared spectra of compound 1.

a,b, Transient UV/vis (a) and infrared (b) difference spectra in MeCN or MeCN-d₃; excitation at 400 nm. c, Calculated UV/vis difference spectra ³MLCT − Cu(I) ground state(GS) (B3LYP/def2-svp) including the qualitative emission spectrum (dashed line). d, Calculated infrared difference spectra ³MLCT − Cu(I) GS of the triplet state (B3LYP-D3/def2-tzvp, PCM).

The amplitudes of the transient difference spectra show a reduction by ~50% during the first 5 ps, after which a decay on the 100 ps timescale takes over. A global analysis of the time-dependent absorption with exponential functions yields four time constants: τ₁ = 0.2, τ₂ = 1.3, τ₃ = 11 and τ₄ = 120 ps. The corresponding fit amplitudes (Supplementary Fig. 2b,d) contain information on the related reactions. Light absorption inducing an optical (d¹⁰–π*) transition populates the electronically excited and symmetry-allowed (singlet) S₁₄ state (obtained by time-dependent DFT (TD-DFT)) (Fig. 3 and Supplementary Section 6). Within our time resolution, the initial reaction dynamics¹¹ lead to the S₁ state, which subsequently relaxes the nuclear coordinates to S_1,relax within τ₁ = 0.2 ps. In the τ₂ = 1.3 ps reaction, the absorption signal decays by ~50%. The shape of the corresponding amplitudes (see Supplementary Fig. 2b) is consistent with a partial Cu(II) to Cu(I) back reaction. After a transient with weak amplitudes (τ₃ = 11 ps), the remaining absorption signal essentially vanishes at τ₄ = 120 ps, re-forming the original Cu(I)-type ground state of 1 in this process. Complementary information on the lifetimes of excited electronic states is obtained from time-resolved optical emission spectroscopy, which reveals a time constant in the 100–200 ps range (Supplementary Fig. 16).

Figure 3: Schematic representation of the involved states.

Assignment by TD-DFT (Supplementary Section 6). The coloured scheme illustrates the angle between the ligand planes in the bis(chelate) complexes.

More details on the nature of the involved states were obtained from transient infrared experiments that probed ligand vibrations (Fig. 2b; width of the instrumental response function, 1.5 ps). The transient difference spectra recorded at the early times suggest that the original ground-state absorption vanishes on excitation, and a new species with Cu(II) character is formed (positive bands at 1,400 and 1,510 cm⁻¹ are directly related to the guanidine vibrational shift because of the Cu(II) character, assigned by DFT (Supplementary Section 4)). Global modelling with three time constants, (τ₂ = ~2, τ₃ = 11 and τ₄ = 120 ps) delivers interesting information about the transient species. The amplitudes (Supplementary Figs 2–9) show that both the τ₂- and the τ₄-related processes are caused by a back reaction from a state with Cu(II) character to a Cu(I) state (for a detailed discussion of τ₂, vide infra). The amplitudes of the τ₃ component show the decay of red shifted absorption combined with a bleach recovery of ground-state bands, that is, features expected for the cooling of a vibrationally hot ground state of 1⁴⁴. Hence, two decay channels exist for S_1,relax: into the hot ground state and into the triplet state T₁ (Fig. 3), which returns with a longer τ₄ constant into the Cu(I) ground state. The analysis of the time-resolved UV/vis and infrared absorption experiments indicate that very small absorption changes persist at later times (>500 ps). Thus, not much more than ~5% of the excited complexes may follow other decay paths, for example, via long-lived triplet states.

Additional measurements with time-resolved optical absorption spectroscopy (Supplementary Section 3) were performed on samples dissolved in dichloromethane (DCM). These results follow the same reaction scheme described above with a slower time constant in DCM, τ₄ = 240 ps, than in CD₃CN, τ₄ = 120 ps. UV/vis experiments with the excitation wavelength λ_exc2 = 320 nm in DCM yielded no systematic differences in the time-dependent absorption changes in the picosecond time range, which indicates no change in the reaction pathway for excitation energies in the 3.1 to 3.87 eV range (Table 1).

Table 1 Overview of all measured decay times for the optical excitation of compound 1.

The time-resolved experiments performed in the infrared and UV/vis regions lead to the molecular reaction scheme in Fig. 3. Light absorption initially populates a high-lying Franck–Condon state S₁₄ with MLCT character, which relaxes rapidly (τ₁ = 0.2 ps) by internal conversion⁴⁵ to a state S_1,relax with the spectroscopic properties: (1) a strong bleach of the original absorption bands of 1, (2) a broad absorption that extends from the blue part of the spectrum to the near infrared, (3) a pronounced absorption band around 500 nm and (4) infrared signatures that show the formation of Cu(II)–guanidine–quinolinyl radical bands. The different spectral features together with DFT calculations support that S_1,relax is the ¹MLCT state with Cu(II) and a quinolinyl radical anion. With a time constant of τ₂ = ~2 ps, there is a strong decay of the excited-state amplitude and the formation of a vibrationally hot Cu(I) species. Its subsequent cooling (11 ps) suggests that about 50% of the population in state S_1,relax decays to a vibrationally hot ground state of 1. The remaining population reacts further to another electronically excited state with Cu(II) character, named T₁, with new properties. A comparison of the amplitude spectra and the calculated infrared-difference spectra for the T₁ state and species 1 (Fig. 2d) strongly suggests that T₁ is, indeed, a ³MLCT state. Moreover, the calculated ultraviolet spectrum for T₁ exhibits ultraviolet absorption at 400 nm (as observed in the transient ultraviolet region) together with a broad absorption at 600 nm, the latter being assigned to transitions within the quinolinyl radical (Fig. 2c). The bifurcation out of S_1,relax towards the hot ground state and the excited-state T₁ is guided by the respective rates 1/τ_2a and 1/τ_2b. The final process for T₁ is its decay (τ_4a = 120 ps) to the reactant state 1. Further long-lived states with a small amplitude observed in the emission experiment can be related to a solvent exciplex T_1,MeCN as discussed in more detail below.

Pump–probe XAS

To obtain more insight into the nature of the ³MLCT state, pump–probe XAS experiments can measure the oxidation state of copper via X-ray absorption near-edge spectroscopy (XANES). In addition, extended X-ray absorption fine structure (EXAFS) spectroscopy is able to determine structural changes around the copper centre^{34,35,36,37,38,39,40}. Copper K-edge XAS spectra of 1 (red), 2 (black) and the optically generated triplet T₁ (green) after 100 ps are shown in Fig. 4a. The measured pumped XAS spectrum, 1_pumped, is a superposition of the ground-state species 1 and the excited-state species T₁ via

where f denotes the photoexcited fraction of excited-state molecules. By scaling the experimental transient-difference spectrum with f⁻¹ to match the static difference spectrum A(2) − A(1), we extract f ≈ 10% (Fig. 4b), as the XANES spectra of 2 and T₁ are very similar. Both species have the absorption edge blueshifted by ~2.5 eV with respect to the XANES of the reactant 1 (Fig. 4a), which is indicative of a copper species of oxidation state +2. Figure 4c expands the pre-edge area where the 1s → 3d + 4p transitions⁴⁶ indicate the 3d⁹ configuration of Cu(II). Both these two observations thus confirm the Cu(II) character of the transiently formed MLCT state³⁷. Furthermore, the XAS spectra of 2 and of T₁ show, for higher energies of about 9,050 eV, a shift of the first resonance peak with respect to the spectrum of reactant 1 (Fig. 4b). This shift is related to the expected structural distortion between Cu(I) and Cu(II) complexes which involves a length contraction of the Cu–N_gua bonds. At energies far above the edge (in the 30–1,000 eV region), the EXAFS spectrum provides information on the interatomic distances next to the type and number of scattering atoms around the selected absorber^47,48,49 (Supplementary Section 4 gives the details). The first coordination shell around copper for the photoexcited complex T₁ is closer than for the reactant 1, by ΔR(Cu–N) = −0.070 ± 0.044 Å from the reactant Cu–N distance of R(Cu–N) = 2.024 ± 0.021 Å, in agreement with our DFT results for both groups of ligand atoms, ΔR(Cu–N_gua) = −0.09 Å and ΔR(Cu–N_qu) = −0.08 Å.

Figure 4: Time-resolved X-ray absorption data.

a, K-edge XAS spectra of 1, 2 and T₁ for an ideal delay. b, Corresponding transient difference spectra 2 − 1 and T₁ − 1. c, Pre-edge region of a. d, XAS transient difference signal between T₁ and 1 as a function of time delay at a fixed energy of 8,984 eV. This energy corresponds to the strongest transient signal as indicated by the vertical line in b. The black curve represents the transient (TR) fluorescence at 510 nm emission up to 2 ns. e,f, Fourier transform (FT) amplitudes of the EXAFS spectra as a function of the radial distance (R) for 1 and T₁. The spectra clearly show the difference in bond distance within the first coordination shell of 1 (e) and T₁ (f), as marked by the dashed vertical lines. k denotes the reciprocal space. In detail, the EXAFS spectra fits of 1 and T₁ reveal a bond-length change of the mean Cu–N distances of −0.070 Å ± 0.044 Å on photoexcitation.

The time dependence of the XAS signal at 8,984 eV, the position of the largest difference to the Cu(I) ground state, is shown in Fig. 4d (red points and error bars). It allows the analysis of the temporal decay of the involved excited states. We also plot the transient fluorescence emission at 510 nm up to 2 ns (black curve). With regard to Table 1, all data sets agree well with each other for the 120 ps feature given the involved error bars. The fit to a model in which the initial decay represents the two depopulation channels of T₁ into 1 with τ_4a and of T₁ into T_1,MeCN with τ_4b is shown in the Supplementary Section 4 and the results are given in Table 1. T_1,MeCN is another MLCT–Cu(II) component of smaller amplitude that decays in the nanosecond time domain, which is related to the formation of an exciplex (Supplementary Section 6 gives the details).

The observed Cu–N bond-length shortening in T₁ occurred because the former highest occupied molecular orbital (HOMO) of 1 had a Cu–N antibonding character⁴² (Supplementary Fig. 19 gives the β-LUMO (lowest unoccupied molecular orbital) of T₁). Excitation of an electron from the HOMO to a quinolinyl π* orbital relieves the antibonding pressure, and thus enables the bond-length decrease.

The structural reorganization at the copper takes place within a few hundred femtoseconds during the formation of the S_1,relax state. As S_1,relax and T₁ are geometrically rather similar and the corresponding spin orbitals of S₁ and T₁ (Supplementary Figs 19 and 21) are twisted against each other, the spin–orbit coupling (SOC) may be relatively large, and hence the ISC is completed within 2 ps, which is fast for an ISC process at copper^11,50. Tahara and co-workers reported ISC values of 8–10 ps for substituted phenanthroline complexes [Cu(dmphen)₂]⁺ (dmphen, 2,9-dimethyl-1,10-phenanthroline) and [Cu(dpphen)₂]⁺ (dpphen, 4,7-diphenyl-1,10-phenanthroline)¹³. Without substitution, no ISC is observed other than a direct transition to the ground state in 1.8 ps (ref. 13). Figure 5 summarizes the observed changes on both reaction coordinates: (1) the angle between the chelate planes and (2) the Cu–N_gua bond length. The reaction path from the Franck–Condon S₁ state over S_1,relax into T₁ benefits from the entatic-state principle because both extremes of the reaction coordinate (1 and 2 in their ground states) are structurally rather similar through the energization by the ligands. The large amount of covalency of the guanidine–Cu bond and the structural constraint lower the degree of potential ‘flattening’ of S_1,relax and T₁ (an implication of the entatic state) and lead to the rotation of the corresponding orbital required for efficient SOC⁵¹.

Figure 5: Visualization of the ‘entatic’ coordinates for the optical excitation of compound 1.

Centre: two-dimensional reaction coordinate of the excitation of 1 and the subsequent decay. One axis depicts the angle between the coordination plane CuN₂ of one chelate ligand versus the coordination plane CuN₂′ of the other chelate ligand (see inset on the right side), hence indicating the degree of torsion between square planar (= 0°) and tetrahedral (= 90°) configuration. GS, ground state. The entatic-state principle applies for the relaxation of S₁ to S_1,relax, to the ISC from S_1,relax to T₁ and to the final electron back-transfer from T₁ to S₀. The red inset illustrates the entatic-state principle for the slice along the diagonal of the diagram. Red inset adapted from ref. 10, Wiley.

T₁, in turn, decays to the ground state rather fast, in 120 ps, by a back transfer of the electron from the formally reduced ligand to the copper ion. This is again at the lower end for ISCs¹¹ and can be traced back to the fact that T₁ geometrically strongly resembles 2 and 1 with an efficient SOC and corresponding orbital rotation (Supplementary Fig. 21). Tahara and co-workers reported values in the nanosecond range for this transition¹³. We propose that this fast ISC can be correlated with the small HOMO–LUMO gap of ~0.37 eV, which is in accordance with the literature⁵². This implies that the entatic-state principle is also applicable for spin-state control. A direct connection between spin-state control and the entatic state has recently been reported for iron enzymes by Solomon and co-workers.⁵³.

Normally, the entatic-state principle is valid for ground-state processes that belong to the classical region of Marcus theory^18,27. Contrastingly, MLCT processes are often Marcus-inverted processes. However, in the present case, the free energy of the T₁ decay is considerably smaller than the reorganization energy (Supplementary Section 6 gives the details). Consequently, the T₁ decay can be treated as a normal Marcus process.

Hence, we propose that the entatic-state principle is not only valid for electron-transfer steps themselves, but also for optically excited charge-transfer processes. As the complexes are restrained by their ligands, their excited states are restrained as well to attain the ground state in due course. This can be regarded as a contrasting effect to that observed in complexes with large substituents, which enforce long triplet lifetimes of up to nanoseconds¹¹.

Conclusions

The energization of Cu(I) and Cu(II) complexes by tailored guanidine–quinoline ligands realizes the entatic-state principle. Herein we document the impact of the entatic-state principle on their photochemical behaviour. The use of complementary experimental techniques allows an understanding of the charge-transfer processes in the copper quinoline–guanidine complexes from the perspective of the electronic system, the structural degrees of freedom of copper and the ligands. Transient absorption and fluorescence spectroscopy deliver all the relevant electronic timescales, whereas transient infrared spectroscopy provides information on structural distortions of the ligand. XAS gives insight into the change of the copper oxidation state and of the first coordination shell. The geometric restraints of the copper(I) quinoline–guanidine complex lead to very fast structural dynamics because all the involved states are structurally similar owing to the pre-organization of the metal coordination by the ligand. This example demonstrates for the first time that the entatic-state principle is highly valuable for photochemistry. To further expand the underlying concept to provide an explanation for a tunable fast spin-state control, new experiments that exploit ultrafast X-ray emission spectroscopy are needed, which can reliably monitor spin states in transition metal complexes.

Methods

All materials and methods are described in detail in the Supplementary Information.

All manipulations that involved air- and moisture-sensitive compounds were performed under pure dinitrogen (N₂), dried over granulate P₄O₁₀, using Schlenk techniques, or in a glovebox with dried solvents. Solvents were either distilled from a sodium benzophenone ketyl radical (THF, C₂H₅OC₂H₅) or from CaH₂ (CH₃CN, CD₃CN, CH₂Cl₂). The complexes were synthesized according to published procedures¹⁰.

The ultrafast absorption changes from the infrared to the ultraviolet were measured in excitation and probe experiments based on a laser-amplifier system (Spitfire Pro, Tsunami, Spectra Physics, repetition rate 1 kHz). Supplementary Section 3 gives the details. Excitation pulses at 400 nm were obtained via a second harmonic generation. Probing pulses in the visible and near ultraviolet were produced by continuum generation in CaF₂. Probing in the mid-infrared used pulses generated by non-collinear and collinear optical parametric amplifiers and difference frequency mixing in an AgGaS₂ nonlinear crystal.

Excitation and properly delayed probing pulses were overlaid in optical cells that contained the sample molecules in solution. The copper compounds were pumped through the cells to exchange the excited volume between two subsequent excitation pulses. The transmission of the sample was recorded using spectrographs in combination with multichannel array detectors (MCT IR-0144, Infrared Systems Development for the mid-infrared and S3902-512Q from Hamamatsu for the ultraviolet and visible regions). The temporal resolution was ~100 fs for the UV/vis spectra and 1.5 ps for the infrared probing experiment. For each delay time position, ~10.000 shots were averaged and the excitation-induced absorption changes were calculated from measurements with and without excitation pulses. The measurements were performed under magic angle conditions and at room temperature.

The transient absorption spectra were globally fitted with exponential functions that led to the decay times τ_k and the decay associated difference spectra (DADS) of the intermediate states.

For the fluorescence experiments, a Tsunami Ti:Sapphire laser system, model HP fs 15 W P, in conjunction with a pulse picker, model 3980-2S (all Spectra Physics Lasers), with an integrated frequency doubler was used. The repetition rate was set to 8 MHz and the laser power in the range up to 10 mW. For time-correlated single-photon counting, a PicoQuant PMT Hybrid 06 detector was used. For the steady-state measurements a thermo-electrically cooled QE65000 spectrometer (Ocean Optics) was used. For the measurements, a Suprasil glass cuvette (Hellma Analytics) was filled with 3 ml of solution. The PMT was connected to a PicoQuant TimeHarp260P single-photon counting PCI express card.

The time-resolved peak-to-peak TRPP XAS experiments were performed at beamline P11 of the PETRA III synchrotron light source at Deutsches Elektronen Synchrotron (DESY). PETRA III was operating in the 40 bunch mode with a time gap of 192 ns between two adjacent bunches and a bunch revolution frequency of 130 kHz. The sample was pumped into an excited state by a femtosecond PHAROS laser (Light Conversion Ltd) at 343 nm and probed by an X-ray pulse.

To prevent radiation damage, a liquid microjet system was used for the sample delivery. The thickness of the jet was between 150 to 300 µm. The temporal overlap of the laser and X-ray pulses was assured via a fast photodiode located at the jet position. To obtain the XAS spectra, the X-ray fluorescence yield mode was used with an avalanche photodiode (ADP) (model APD001 from FMB Oxford) in a 90° configuration. A second APD for normalization was directed towards a metal foil located several centimetres behind the sample position. The acquisition rate was set to 130 kHz, which is twice the repetition rate of the laser system of 65 kHz. The used digitizer (Model ADQ412AC, 12 bit, 2/4GS (SP Devices)) was allowed to register multiphoton events and thus drastically improve the signal-to-noise ratio.

For the quantum chemical calculations of the excited states, B3LYP/def2-svp was used as density functional method. The key geometric data of the symmetry-allowed singlet excitations are summarized in Supplementary Table 5. The reorganization energy of the ISC can be described as the sum of the reorganization energies of the corresponding S₀ and T₁ ion pair. The reorganization energy of each cation can be calculated by its optimized ground-state energy and the single-point energy of, for example, a singlet configuration in the ligand environment of the corresponding T₁ complex (denoted as SingL(Trip)) (Supplementary equation (4) and Supplementary Fig. 23).

Data availability

The data that support the findings of this study are available from the corresponding author on request.