Multifaceted Deactivation Dynamics of Fe(II) N-Heterocyclic Carbene Photosensitizers

Excited state dynamics of three iron(II) carbene complexes that serve as prototype Earth-abundant photosensitizers were investigated by ultrafast optical spectroscopy. Significant differences in the dynamics between the investigated complexes down to femtosecond time scales are used to characterize fundamental differences in the depopulation of triplet metal-to-ligand charge-transfer (3MLCT) excited states in the presence of energetically accessible triplet metal-centered (3MC) states. Novel insights into the full deactivation cascades of the investigated complexes include evidence of the need to revise the deactivation model for a prominent iron carbene prototype complex, a refined understanding of complex 3MC dynamics, and a quantitative discrimination between activated and barrierless deactivation steps along the 3MLCT → 3MC → 1GS path. Overall, the study provides an improved understanding of photophysical limitations and opportunities for the use of iron(II)-based photosensitizers in photochemical applications.


Artefact evaluation
We have utilized thin cuvettes (100 um path length) with ultrathin windows (200 um).With that, we have drastically reduced the contribution of the cuvette and solvent ("solvent response") to the solute response both in amplitude and in time.The solvent response (i.e. the transient absorption (TA) signal of the thin cuvette with solvent) is shown in Figure SI.1.Note that the strongest signals have a peak amplitude of ~0.1 mOD.The weakest TA signals displayed in this manuscript is on the order of ~0.5 mOD at the early peak even though attenuation of the excitation due to absorption of the solute is not taken into account for these measurements.In the red part of the TA spectrum (550-800 nm) where resolving the ultrafast decay component is crucial for this study, the solvent response is nearly twice weaker than in the blue, see Figure SI.2.Therefore, we state that the ultrafast dynamics of the red excited state absorption (ESA) in complex 1 is due to processes in the solute, not artefacts caused by the solvent or cuvette.All TA data shown was measured with a perpendicular polarisation between pump and probe beams with a Glan-Thompson polarizer in the probe set to block scattering of the pump beam.As a consequence of this, we avoid pump scatter at the excitation wavelength (470 nm, 495 nm resp.515 nm), but instead in the time-resolved spectra we see a contribution of the pump spectrum in the pump-probe pulse overlap.We associate this signal to the probe-induced Kerr effect for the pump beam in the sample (see Figure SI.3), which however does not affect the conclusions we make regarding the ESA evolution.

UV-TA data
Here we present the measured TA data of complex 1 and 2 in the UV spectral range (280-380 nm).
The TA data was fitted by global analysis using available software. 1 To avoid the early time artifact contribution, the data was cut before 1 ps and the instrument response function (IRF) was set as 90 fs.For TA data in the UV range of complex 3, we refer the reader to the publication by Duchanois et al..

Vis-TA data
To further understand what processes return population to the ground state in both molecules, single kinetics from the GSB bands were fitted by a sum of exponential functions.The fitting was conducted from ~100 fs or later to exclude rise components.For 1, TA dynamics in the GSB region

GS cooling fits
The spectral shape of the feature that correspond to GS cooling in the TA data should consist of a combination of negative steady-state absorption and a broadened (i.e.hot), positive ground state absorption.Here we present modelling of the TA spectra of 1 at late (≥30 ps) delay times, where this component should be prominent whereas less than 3% of the faster components should remain.As a first approximation, the lowest energy peak of the GSB band was fitted by a gaussian function.This negative gaussian function was summed up with a positive broader gaussian with the same area, to approximate the ground state absorption.The TA spectrum around the red GSB edge at different spectral ranges and delay times were fitted by this sum, see Figures SI.14

Oscillations
The TA datasets of all complexes feature oscillations, mainly in the ESA wavelength range.Here we show a fit of the oscillations at the kinetics where they are the most prominent.Oscillations are identified by looking for oscillatory features in the residuals of the fit.The number of oscillatory components added was determined until the residuals were random (see Figures SI.16-18).In Table SI.3, the parameters of the fitted oscillations are collected.

Target analysis
The different models considered in the target analysis are depicted here, together with their names and naming of the fitted states and decay rates."br" stands for branching and is the faction of population that would go into the indicated state, defined so that the total population in the system is always 1.The TA data is fitted starting from 300 fs, to ensure that we only quantify processes that are slower than IRF in our experiment.The evaluation of the applicability of a model is based on the quality of the fit, randomness of the residuals and interpretability of the species associated spectra (SAS).To further quantify the disagreement of the model in Kunnus et al. 4 with our slow timescale data we applied target analysis using their model as the target.Since the branching between the 3 MLCT and the 3 MC states proposed in their paper is largely over at 300 fs, we introduce the branching ratio in the analysis as a varied parameter.First, we applied the model "NatComm" (see  Based on the assignment discussed in the main manuscript that complex 1 first undergoes ultrafast MLCT->MC conversion (where target analysis would not be reliable due to the IRF response), two models for the subsequent dynamics were found to yield results that met all evaluation criteria.First, a purely consecutive model We were also able to fit the TA with a parallel decay scheme containing two independent states between which the population is initially split, and with one of the states featuring a relaxation/cooling process such that this model also contained three separate rates (see Figure SI.19 "parallel ").This model yielded time components 610 fs, 13 ps and 20 ps, see associated SAS in Figure SI.35-37.It is noteworthy that the similarity of these rates with those from the consecutive model discussed above suggest that the TA data clearly require these characteristic time components to be present even if different detailed deactivation schemes can be used to fit the data with similar accuracy.This also meant that testing revealed that there were some residual ambiguities both in the initial branching ratio and such that the target analysis could not distinguish whether the 610 fs relaxation/cooling should be associated with the 13 ps or 20 ps state meaning that the 610 fs SAS changes shape according to the state that follows in the kinetic model (see Figure SI.36 and SI.37).The branching ratio between the different states is coupled to the relative amplitudes of the decaying components and thus could not be evaluated based on this type of analysis.To summarize, the key aspect of this model is that it involves the simultaneous initial population of two 3 MC states (yielding ESA B resp.C) which decay independently to the GS with lifetimes of 13 ps and 20 ps, respectively, after some initial Here the lifetimes of the 3MLCT and 3MC states were locked according to the values in the paper, however cooling of the 3MLCT state and GS are freely fitted.The branching ratio is set to 88%, in order to yield equal extinction of GSB in all SAS.The SAS of 3MLCT* and 3MLCT are similar which is reasonable.The 3MC SAS is also very similar in shape to the 3MLCT SAS but red-shifted.The ESA spectra associated with each state are of course not known, but it seems unlikely that they would be that similar.The GS cooling SAS still assumes a shape that we cannot explain by a broadened GSA together summed with the GSB signal.

Results from NatComm model
The errors are similar to the other best models, even though some components were locked.Here however, there is one more component in the fit compared to the other best results that can compensate for the locked components.All-in-all, the need for still locking components and adding one more compared to other models together with the SAS shapes rules this model out.A purely consecutive model with 3 components, here in order from shortest to longest.This is the only order that works without yielding unphysical SAS with negative features in the ESA spectral range.This is one of the fits yielding lowest residuals and acceptable SAS.Nevertheless, there are fits with branching with less residuals, but they also have more fitted parameters.Here the branching ratio was set to 42% based on yielding similar GSB extinction in all SAS.Since the SAS do not show the exact same GSB feature, it is not possible to scale the branching ratio by matching the GSB extinctions therefore we cannot make any conclusions about it from our data.

Anisotropy
The anisotropy signal was calculated by subtracting the perpendicular dataset from the magic angle dataset, and then dividing with the magic angle dataset.Special attention was paid to the single kinetics, and to double check that the background-and chirp corrections were good enough, see the data plotted in Figure SI.38.Since still the relative time zero between MA and L kinetics could differ on <100 fs precision, anisotropy kinetics were only evaluated after this time.Furthermore, when the TA signals approach low values at >30 ps delays the signal-to-noise of the anisotropy signal degrades.Therefore, the anisotropy kinetics cannot be evaluated at much later delay times than ~30 ps.To be in a reliable time-range with high enough signal-to-noise and minimum effect from potential errors in the chirp-correction, anisotropy spectra were plotted from 300 fs to 20 ps see Figure SI.39.In the anisotropy signal of complex 1, we have seen that the ground state bleach anisotropy is not changing much within this time window.This means that the observed change of anisotropy in other spectral regions is not due to reorientation dynamics of the molecule and solvent, which also agrees with the longer reorientation time expected for rather bulky transition metal complexes (see for example Malone et al. and Wallin et al.). 5,6Since all complexes share a similar molecular structure, we expect that also 2 and 3 will not show reorientation dynamics on the sub-20 ps timescale.Therefore, we consider it safe to look at perpendicular polarisation TA data for the population dynamics on sub-20 ps time scale in all complexes.In Figure SI.39, selected anisotropy spectra of 1 are shown together with the TA spectrum at 10 ps.Interestingly, in the part of GSB not contaminated by scattering pump set at magic angle the anisotropy is nearly the same at all delay times.This confirms that the change of anisotropy in other wavelength ranges should be related to state-to-state transitions rather than to reorientation dynamics. 5,6It is important to note that a hot GSA (considered as a potential assignment of the ESA C feature) should exhibit anisotropy close to the rest of the GSA range as the dipole moment orientation of the "hot" and thermalized S must be the same.In the case of GS cooling, all competing ESA should largely vanish at this stage of energy relaxation and the conversion of the 3 C state into a "hot" S should lead to an increase of anisotropy contrary to the observation.In the anisotropy kinetic at 530 nm, we fit instead a decay of 29 ps clearly different to the fitted biexponential 230 fs and 13 ps decay at 660 nm see Figure SI.40.All these considerations together argue that the 16 ps ESA C component cannot be dominated by a GS cooling process, but rather reflects an excited state decaying to GS.

Temperature dependent TA fits
In this section we show the temperature dependent TA data fits by global analysis using the available software. 1 To avoid the early time artifact contribution, the data was cut away before 500 fs and the instrument response function (IRF) was set as 90 fs.The data was fitted in the ESA wavelength region by a single exponential function to get the major excited state lifetime required for the Arrhenius analysis, see Figures SI.41-46.The GSB region was not included in the fit due to the large pump scattering in this region.where   is the excited state lifetime,  0 is the temperature-independent deactivation rate,  is the pre-exponential factor, ∆ is the energy barrier,  is the Boltzmann constant and  is the temperature.

Stability
In Figure SI.48, we show the absorption spectrum for the samples used for TA measurements.This is a way for securing the stability of the sample, why the absorption after the measurement is also checked.As is evident from Figure SI.48, the absorption spectra are similar in all cases, except for the concentration.It is common that the concentration can increase during the measurement, due to solvent evaporation.This proves that the sample quality was good in all measurements, and even if case some part of the sample had degraded this could not have been excited by the excitation light (470 nm) as no new absorption was observed in the excitation wavelength region.) 1000/T (K -1 ) T=139 K

Figure SI. 1 .
Figure SI.1.Solvent response dynamics measured at identical conditions as the measured TA data of a) the acidic buffer in the thin cuvette used in the measurements of complex 3 and b) acetonitrile in the thin cuvette used in the measurements of complexes 1 and 2.

Figure SI. 2 .
Figure SI.2.Solvent response compared with the TA signal for complex 1 in acetonitrile at a) 485 nm and b) 650 nm.The two graphs show two different datasets where the concentration of 1 was slightly different.The solvent response was scaled to account for similar pump light absorption as in the low concentration sample.

Figure SI. 3 .
Figure SI.3.Normalized TA spectra at 50 fs delay time for all investigated complexes showing the distorted ground state bleach (GSB) spectra by the pump spectrum at the wavelength indicated by vertical lines.This is assigned to the Kerr effect.The contribution of this signal has been cut in the data shown in the main manuscript.Steady state absorption of each complex is shown by the dashed lines.No decay associated spectra (DAS) for the component faster than the instrument response function (IRF) is shown in the main manuscript, as the shape of any component on these short timescales could be substantially influenced by the uncertainty in the IRF.In Figure SI.4,we show a comparison between the fastest DAS and the TA spectra measured at 50 fs delay in the ESA wavelength range as this is the range where we have established the contribution of this component.

Figure SI. 4 .
FigureSI.4.Fitted decay associated spectra of complex 1.The decay associated spectra with time component 50 fs is compared to the TA spectra measured at 50 fs delay time.

Figure SI. 5 .
Figure SI.5.TA spectra of a) 1 and b) 2 in the UV wavelength range.Spectra are chirp-and background corrected.The (inverted) linear absorption spectra of each molecule are included for comparison.

Figure SI. 6 .
Figure SI.6.TA kinetics of a) 1 and b) 2 in the UV wavelength range.TA data is chirp-and background corrected.Fits are represented by the solid lines.

Figure
Figure SI.7.Decay associated spectra of TA data of a) 1 and b) 2 in the UV wavelength range.The (inverted) linear absorption spectra of each molecule are included for comparison.

Figure SI. 8 .
Figure SI.8.TA spectra of 1 in the UV to NIR wavelength range.The relative scale of the UV and vis-NIR TA data is arbitrary, set by the authors.The black line indicates the transition between the two datasets, the dotted black line -the inverted steady-state absorption.TA spectrum at 50 fs delay time is only shown in the red wavelength range.

1 .
In Figure SI.11 we show for clarity the ESA wavelength range of the earliest measured TA spectra of complex The ESA B feature is first clearly seen at 100 fs delay time, when ESA feature A has decayed to some extent.The ESA B feature then shifts from 540 nm (at 100 fs) to 520 nm (at 1 ps).

Figure
Figure SI.11.Early TA spectral evolution of complex 1.The plot is displaying the ESA wavelength range in order to follow the shift of spectral feature B. Spectra are chirp-and background corrected.The inverted steady-state absorption is indicated by the dashed black line.

Figure SI. 12 .Figure
Figure SI.12.Global fit with two components in the GSB region (480-540 nm) of 3. Data was chirpand background corrected, fit performed from 300 fs.
-15.Different fit results can be found depending on what wavelength range is included in the fit, see Figure SI.14.None of the fits reproduce the full spectral shape satisfactorily.

Figure
Figure SI.14.TA spectra of 1 at 40 ps delay time, fitted with the GS cooling model.Here three different cases are compared, where the data has been cut at different places.

Figure
Figure SI.15.TA spectra of 1 at a) 30 ps delay time and b) 50 ps delay time, fitted with the GS cooling model.

Figure
Figure SI.16.a) TA kinetic at 525 nm of 1, selected for having strong oscillations, with fit.b) Residuals after the fit.

Figure
Figure SI.17.a) TA kinetic at 550 nm of 2, selected for having strong oscillations, with fit.b) Residuals after the fit.

Figure
Figure SI.18.a) TA kinetic at 650 nm of 3, selected for having strong oscillations, with fit.b) Residuals after the fit.

Figure
Figure SI.19.The different models analysed by target analysis.
Figure SI.19)without any change with locked3 MLCT lifetime of 9 ps,3 MC lifetime of 1.5 ps and branching ratio of 60:40.4This model failed to fit the data due to non-random residuals and we conclude that our data disagree with the previously proposed model (see Figure SI.20).Next, we consider what modifications of the model can be made to tune it to fit the data.First, with freely varied decay times the fit naturally results in a substantial decrease of errors but the lifetimes changed to 610 fs, 13 ps and 20 ps.Furthermore, in the ESA wavelength range the resulting SAS exhibits a negative feature, which is an unphysical result as no species could have a contribution of a negative sign in this spectral region (see FigureSI.22).We then further adapted the model by adding one cooling/relaxation step in either the 3 MLCT or the3 MC state (see Figure SI.19).For all attempts it was however still hard to interpret the fitted SAS in a physically meaningful way (see Figures SI.20-31).The best possible fitting required drastic variation of the branching between the 3 MLCT and the 3 MC with >75% of the population going to 3 MLCT, together with locking the lifetimes to the values from Kunnus et al. 4 (see Figure SI.27).

Figure SI. 20 .
Figure SI.20.The model from the Nature Comm.paper, with fixed MC and MLCT lifetimes as well as the selected branching of 60% according to the publication.This fit results in clear non-random residuals and was therefore ruled out.
FigureSI.28.The exact same model, but with the 60% branching as specified in the Nature Comm.paper.With this branching the SAS yields unphysical negative features in the ESA spectral range.The model must therefore be ruled out.Setting the branching ratio to more than 75% was found as the limit for avoiding the unphysical negative SAS.
Figure SI.33.Again, a consecutive model with the similar 3 components, however in another order.This resulted in one unphysical SAS showing negative features in the ESA spectral range.Therefore, the model was ruled out.

Figure
Figure SI.38.TA kinetics measured at magic angle (MA) or perpendicular (L) polarisation between the pump and the probe beam used for constructing the anisotropy kinetics.a) The 430 nm kinetic, b) the 530 nm kinetic and c) the 660 nm kinetic.Data was corrected for chirp and background.

Figure
FigureSI.39.Anisotropy spectra of 1 at selected delay times, cut to remove excitation scatter and corrected for background and chirp.In dashed line the TA spectrum at 10 ps is shown for comparison.

FigureFigure
Figure SI.41.Single exponential fit of the TA data of 1 at room temperature, shown are the fitted kinetics.

Figure
Figure SI.43.Single exponential fit of the TA data of 1 at 180 K, shown are the fitted kinetics.

Figure
Figure SI.44.Single exponential fit of the TA data of 1 at 160 K, shown are the fitted kinetics.

Figure
Figure SI.46.Single exponential fit of the TA data of 1 at 120 K, shown are the fitted kinetics.Due to the bad signal-to-noise at this low temperature, a non-decaying component had to be added to account for the background offset).

Figure
Figure SI.47.Temperature dependence of the major excited state lifetime in 1, with fitted Arrhenius expression.The glass transition temperature of butyronitrile (139 K) is marked by a vertical line.

Figure SI. 48 .
Figure SI.48.Absorption spectra of samples used for three different TA measurements before and after TA.As a further line in assuring the quality of the measured samples, we also compare the quantified TA data measured by us and in Liu et al. and Kunnus et al..In all three papers, the major decay component found was 9 ps.The decay associated spectrum associated with this component also looks similar in all three papers, with a peak ~530 nm (2.3 eV).Furthermore, all papers have

Table SI .1. Fit components for the single kinetic in the GSB region of 1. Wavelength (nm) Comp (fs) Comp (ps) Comp (ps) 380
A decay component shorter than IRF is also needed for the fit here.This could be due to the Kerr effect, as these wavelengths are close to the excitation wavelength 470 nm.It could also be the rise of ESA features.
*Figure SI.10.GSB kinetics of 3 in the visible wavelength range.The black lines represent the fits that are also shown in the Table below.Table SI.2.Fit components for the single kinetic in the GSB region of 3.

Table SI .
3. Table concluding the oscillation fits.
3"consecutive" FigureSI.19)wastestedandfoundto fit the data with three decay components of 630 fs, 9 ps and 17 ps.This consecutive model only produces meaningful SAS when the rates are ordered from fastest to slowest process (see FigureSI.32-34).Furthermore, the similarity of the 630 fs and 9 ps SAS strongly suggests that both these decay rates are associated with the same state.The SAS belonging to the 17 ps lifetime instead shows signatures of having a separate origin.Summarizing the connection of the spectral features with the different steps in this consecutive model yields the following description of the dynamics as illustrated in the Figure SI.19 "consecutive".The initially populated 3MLCT state characterized by broad ESA in the red quickly (within IRF) converts into a hot3MC state with the spectral characteristics of feature B. This state undergoes sub-ps cooling (630 fs), and then decays on 9 ps to a second 3 C state (3 C'), before this second 3MC state finally decays back to the 1GS on a 17 ps timescale.