Parallel triplet formation pathways in a singlet fission material

Synthesis

As shown in Fig. 1a, HR-TDPP-TEG was synthesized via Sonogashira carbon-carbon coupling reaction. The reaction involves acetylene functionalized HR and respective di-bromo derivatives of TDPP-TEG (TDPP-TEG-Br) in the presence of palladium catalyst Pd(PPh₃)₄ in tetrahydrofuran (THF) and diisopropylamine (DIPA) mixture. The HR was synthesized according to the reported procedure with necessary modifications⁴⁵. A detailed synthetic procedure and structural characterization are shown in Supplementary Note 1 and Figs. 1–4.

a TA spectra of HR-TDPP-TEG in THF after photoexcitation at 2.3 eV with 35.4 μJ/cm² at room temperature. Steady-state absorption (gray solid line) and PL (gray dashed line) are presented for reference. b averaged TA kinetics (open symbols) in the two spectral regions (blue—SE region, red—PIA region). The black solid lines are the best global fit of the two kinetics with a triexponential function.

a TA spectra of film 1 after photoexcitation at 2.3 eV from 0.2 ps to 1 ns with 1.75 μJ/cm². Steady-state absorption (gray solid line) and PL (gray dashed line) are presented in (a) for reference. TA spectra near 2.3 eV are truncated due to pump scattering. b equivalent to (a) but divided into three time regimes to highlight isosbestic points (gray vertical dashed lines) involved in each step of the spectral evolution. The bottom panel is taken from excitation with 53.1 μJ/cm² to show the presence of the same isosbestic point as in the 1.75 μJ/cm² data. c TA spectra in the range of 1000–7000 ps with the pump fluence of 53.1 μJ/cm². The first derivative of the steady-state absorption spectrum (black solid line), which is shifted in parallel, is also shown for comparison. d Low- (1.75 μJ/cm² for 2.3 eV pump and 3.5 μJ/cm² for 1.84 eV) and high-pump-fluence (53.1 μJ/cm², 2.3 eV pump) TA kinetics averaged in the range of 1.5–1.35 eV. The black lines are the best global fits of the 2.3 eV pump kinetics with a multiexponential function.

a TGPL map of film 1 following photoexcitation at 2.1 eV with a pump fluence of ∼50 μJ/cm². b Comparison of TGPL (integrated PL) and TA (averaged PIA) kinetics with different pump fluences. The time constants in the panel were obtained by fitting the 50 μJ/cm² integrated PL kinetics (yellow line) with a multiexponential function. c Comparison of normalized PL spectra (light color—raw, dark color—smoothed) averaged over two different time regions (1–10 ps vs. 10–100 ps) in the data measured with 50 μJ/cm². The inset shows the difference in peak positions between the two spectra.

Steady-state optical characterization

In dilute THF, the absorption spectrum exhibits a vibronic progression with a spacing of 0.18 eV (Supplementary Fig. 5), which signifies that the S₀–S₁ transition of HR-TDPP-TEG is coupled with a C=C stretching mode. The observed 0-0 peak position at 2.02 eV is substantially red-shifted from TDPP-TEG (2.27 eV), thanks to the extended conjugation afforded by the HRs^33,46. This behavior can be further seen in the HOMO and LUMO electronic distributions calculated by DFT (Fig. 1b), which dominate the S₁ character. The photoluminescence (PL) spectrum is distinctly narrower with different vibronic structure, indicating that the monomer S₁ state is also coupled with a low-frequency torsional mode (typically less than ω = 15 meV) and that structural relaxation along that vibrational coordinate reduces conformational disorder in S₁⁴⁷. This overall behavior is consistent with other monomeric DPPs, indicating that our functionalization does not strongly change the photophysics of isolated molecules.

We prepared thin films of HR-TDDP-TEG via drop-casting, obtaining turquoise-colored film 1 from chlorinated solvents (4:1 mixture chloroform (CF): chlorobenzene (CB) or pure CF) and dark-blue-colored film 2 from THF (Fig. 1d). AFM measurements reveal that these film types correspond to completely different microscopic morphologies (Fig. 1d). The film prepared from chlorinated solvents was composed of elongated fibrillar structures with a mean diameter of 50 nm (Fig. 1d, left). In sharp contrast, highly dense ribbon-like microstructures are observed in the film made by THF (Fig. 1c, right).

These different structural motifs are reflected in different intermolecular interactions, as determined from the optical spectra. Compared to the monomer, the steady-state absorption of film 1 exhibits a strongly enhanced ratio of the 0-0 peak to the 0-1 peak intensities: R_abs = A₁/A₂ ∼ 1.68 in film 1, versus 1.11 in the monomer. Film 2, on the other hand, reveals the opposite behavior, namely relative suppression of the 0-0 peak, with R_abs ∼ 0.69 (Supplementary Fig. 7). Such peak ratio changes are characteristic of different modes of exciton coupling, and we tentatively assign film 1 as a J-like aggregate and film 2 as an H-like aggregate. To estimate the inter-site coupling strength in thin films, we have analyzed the R_abs values in more detail based on the method of Spano and co-workers (Supplementary Note 2)^48,49. R_abs is directly related to the exciton bandwidth (W) which in turn yields the exciton coupling strength J₀ as W = 4|J₀|, as long as we restrict to nearest-neighbor coupling and assume that the nature of coupling is purely Coulombic. With the Huang-Rhys factor (λ²) and vibrational frequency (ω_vib) from the monomer spectra (Supplementary Fig. 6), W values are estimated to be 80 meV for both film 1 and film 2. These values are found to be smaller than the nuclear reorganization energy due to this vibration (λ²ω_vib) of 166 meV, indicating that the HR-TDPP-TEG molecules in thin films are only weakly excitonically coupled. Nonetheless, the spectral shifts for the two films in relation to the monomer are large (120–130 meV, Fig. 1b and Supplementary Fig. 5) and toward the lower energies, in contrast to the expected blue-shift for an H-like aggregate. This effect is due to the gas-to-crystal shift, which here outweighs exciton coupling effects^49,50,51. We verify this picture with our simulated absorption and PL spectra using a Frenkel-Holstein Hamiltonian based on the parameters described above and in Supplementary Table 1. This method yields excellent agreement with the experimental spectra, confirming our analysis from R_abs. The net result is that the opposite modes of coupling—J-type versus H-type—yield coupled excitonic states that are very similar in energy.

The PL spectrum of film 1 reveals clear vibronic emission with a small Stokes shift and resembles a mirror image of absorption (Fig. 1b, middle). This is in line with typical spectral characteristics of J-aggregates⁴⁸. Nevertheless, given that the equivalent vibronic peak ratio R_PL (I^0-0/I^0-1) is smaller in film 1 than in the monomer (Fig. 1b), it can be deduced that the lowest exciton has a localized nature. This result is unsurprising for a system in the weak exciton coupling regime. In contrast, film 2 exhibits energetically broad PL with a large Stokes shift (the peak maximum cannot be precisely estimated due to the detection limit of our ICCD detector). Such a drastic shift implies that film 2 accesses a more relaxed state from the lower Frenkel state, which can be tentatively assigned to the excimer-like state that is widely observed in H-aggregates of various organic materials^{52,53,54,55,56,57,58}. The PL quantum yields of both films are very low, <1%, compared to 59% in the monomer, suggesting the presence of a fast non-radiative deactivation channel from the emitting state.

Excited-state dynamics of HR-TDPP-TEG monomer

To understand these non-radiative decay channels, we turn to femtosecond TA spectroscopy. As a baseline for intramolecular photophysics, we present in Fig. 2 the results for HR-TDPP-TEG in THF. After photoexcitation at 2.3 eV in THF solution, we resolve positive ground-state bleach (GSB) and stimulated emission (SE) signals between 2.4 and 1.7 eV and negative photoinduced absorption (PIA) bands above 2.4 eV and below 1.7 eV (Fig. 2a). As delay time increases, the SE bands dynamically redshift, and this corresponds with a slight change of shape of the PIA. This spectral evolution can be ascribed to structural relaxation towards the S₁ potential minimum along the torsional coordinate, which occurs biexponentially with the time constants of 1 and 74 ps (Fig. 2b). The S₁-state lifetime is estimated to be 2.75 ns, in good agreement with transient PL experiments using the ICCD (2.83 ns, Supplementary Fig. 8). We could not find any triplet signature from monomeric HR-TDPP-TEG (Supplementary Fig. 9), which is in line with its high PL quantum yield (59%) as well as typically low triplet yield (<1%) of other DPP derivatives in the solution phase³³.

Triplet-like photoinduced absorption

Figure 3a, b shows the TA spectra for film 1 excited at 2.3 eV with a pump fluence of 1.75 μJ/cm², a level at which singlet-singlet annihilation effects were found to be negligible (Supplementary Fig. 10). Figure 3c highlights the long-time spectral evolution (>1 ns) following excitation at a higher fluence of 53.1 μJ/cm² to improve the signal to noise. The same features are present in the low-fluence experiment but cannot be so clearly discerned (Supplementary Fig. 11).

Immediately after photoexcitation, the transient response in film 1 is dominated by sharp GSB and SE features in the 2.2–1.6 eV region that match the steady-state absorption and PL spectra. These are accompanied by PIA bands towards the edges of our spectral range (Fig. 3a, b). These initial signatures can be readily assigned to the lower Frenkel state due to the presence of well-structured SE. On long timescales (Fig. 3c), the weak residual signal carries a series of derivative-like features in the ground-state absorption region as well as a clear PIA in the 1.7–1.3 eV range. This latter band, though rarely characterized, is a unique marker of triplets in TDPP derivatives^34,59,60, and its presence demonstrates sub-nanosecond (ns) triplet formation. From the TA data alone, it is not possible to determine conclusively whether this state arises from SO-ISC (one triplet exciton) or SF (correlated triplet pair or two triplet excitons). However, on the basis of temperature-dependent PL measurements (see below) we can assign the TA spectra detected after 1 ns to the spin-0 correlated triplet-pair, i.e., ¹(TT) state, which is the immediate product of SF. We additionally confirmed by ns-TA that the identical triplet-like PIA in the near-infrared region persists up to microsecond timescales (Fig. 3c), supporting the validity of the assignments above. The detailed discussion on ns-TA data for film 1 can be found in Supplementary Note 3. We also explicitly highlight that we detect spectral evolution between only two electronic species, identified as either lower Frenkel or triplet-related states (¹(TT) or T₁), over the entire experimental time window (Fig. 3 and Supplementary Fig. 12). This strongly implies that the direct population of other electronic species, e.g., charge-separated (radical anion and cation) states, in the excited-state dynamics in film 1 can be neglected.

The sub-ns spectral evolution to form ¹(TT) is surprisingly rich, exhibiting complex multiexponential behavior rather than a simple mono-exponential conversion between the two states. The dynamics recorded at the principal PIA band from 1.5 to 1.35 eV can be described with a bare minimum of four-time constants: 7, 43, 300, and 1200 ps (Fig. 3d). Only the final time constant has an immediately evident interpretation, as the decay lifetime of ¹(TT). To probe the role of excess pump energy to the early-time dynamics, we repeated the experiment using red-edge excitation (1.84 eV) also at low fluence (3.5 μJ/cm²). We obtain a perfect overlap of the kinetics over this wide range of excitation energies (Fig. 3d), revealing that vibrational relaxation or cooling processes make a negligible contribution to the kinetics.

Our kinetic analysis suggests distinct spectral evolution processes prior to 1 ns, and indeed, in each temporal range we detect a unique set of isosbestic points offset from the baseline (Fig. 3b). Such points are a spectroscopic signature of population transfer between two states having different electronic nature. In the initial evolution (7-ps kinetics, 0.2–10 ps panel in Fig. 3b), the GSB intensity barely changes but the SE is strongly quenched. At the same time, the PIA at 1.5 eV from ¹(TT) starts to emerge. We thus interpret this process as ¹(TT) formation, but the presence of SE beyond the 7-ps window tells us that not all population from the lower Frenkel state undergoes fast ¹(TT) formation. In the second and the third processes (43- and 300-ps kinetics, 10–200 ps and 200–1000 ps panels in Fig. 3b), although we still detect isosbestic points offset from the baseline, their positions are slightly shifted, and they are accompanied by a simultaneous drop in the GSB intensity. This combination of features strongly suggests that population transfer from the lower Frenkel state to ¹(TT) still occurs but in parallel with significant population loss to the ground state. This behavior would be characteristic of different sub-ensembles within the film showing weaker electronic coupling, and thus slower ¹(TT) formation that is in competition with direct decay from the lower Frenkel state. Overall, the TA data suggest that film 1 undergoes heterogeneous ¹(TT) formation (forming ¹(TT) with different rates) resulting from the microscopic morphological inhomogeneity of thin films.

Transient grating PL (TGPL) results for film 1 provide further insight into the population kinetics of the lower Frenkel state (Fig. 4). To secure a sufficient signal-to-noise ratio of the data, we performed the experiments with a high-pump fluence of about 50 μJ/cm² and confirmed that the integrated PL kinetics are in line with the averaged PIA kinetics in TA obtained with the same pump fluence (Fig. 4b), suggesting that we unambiguously track the kinetics of the same excited-state species, i.e., lower Frenkel. The higher pump fluence data (240 μJ/cm²) additionally support that the sub-ps dynamics stem from singlet-singlet annihilation. After the singlet annihilation, biexponential decays with 8 and 47 ps were obtained through the multiexponential fitting, which is in good agreement with the time constants observed in the TA data. The slower component (300 ps) was not detectable due to the limit of the delay stage. During the 8- and 47-ps decays, we observe no intensity rise in the kinetics anywhere in our detection region. This allows us to rule out relaxation along the same exciton manifold, which typically presents as a dynamic redshift in time-resolved PL spectra with a decay at higher energy and a rise at lower energy⁶¹. We note, however, that the peak positions (∼10 meV difference) and vibronic peak ratios on these characteristic timescales clearly differ (Fig. 4c), implying a slightly different electronic structure between the two. Considering as well the substantial quenching in emission intensity that these decays reveal, and the close agreement with TA kinetics, we find that initial time constants reflect parallel ¹(TT) formation from a distribution of lower Frenkel states.

We have carried out equivalent TA experiments on H-like film 2 as well. In contrast to film 1, even at the earliest timescales, we observe no SE bands, consistent with the weak emitting properties of the H-type coupled state (Fig. 5a). The response is instead dominated by strong GSB and PIA bands and bears a resemblance to the behavior of other organic molecules forming H-aggregates and showing excimer-like states^55,58. Also unlike film 1, we detect no significant spectral evolution or isosbestic points over the full decay range. Although film 2 also reveals long-lived signals into the nanosecond regime, no PIA features are in the NIR region (Fig. 5b), only a surprising positive band (see below). Using a similar analysis of the kinetics averaged from 1.8 to 1.35 eV, we obtain a multiexponential decay with time constants of 12, 84, and 363 ps. Given the lack of spectral evolution, we speculate that the transition from the initial Franck-Condon state to the excimer-like state is faster than the IRF of our setup (200 fs), as other excimer-forming systems, because the monomers are already stacked in the ground state⁵⁸. A detailed spectral analysis of film 2 is presented in Supplementary Fig. 12. Crucially, there is no evidence of any timescale of the triplet-like PIA in film 2, implying that the ¹(TT) formation channel is blocked. Furthermore, we were not able to obtain TGPL signals from film 2, which is presumably due to the very weakly emitting nature of the excimer-like state.

a TA spectra of film 2 after photoexcitation at 2.3 eV from 0.2 ps to 1 ns with 1.75 μJ/cm². Steady-state absorption (gray solid line) and PL (gray dashed line) are presented in a for reference. TA spectra near 2.3 eV are truncated due to pump scattering. b TA spectra in the range of 1000–7000 ps with the pump fluence of 53.1 μJ/cm². The first derivative of the steady-state absorption spectrum (black solid line), which is shifted in parallel, is also shown for comparison. c Low- (1.75 μJ/cm² for 2.3 eV pump and 3.5 μJ/cm² for 1.84 eV) and high-pump-fluence (53.1 μJ/cm², 2.3 eV pump) TA kinetics averaged in the range of 1.8–1.35 eV. The black lines are the best global fits of the two 2.3 eV pump kinetics with a multiexponential function.

Photoinduced thermal artifacts

We note that both films exhibit derivative-like TA spectra in the region of the ground-state absorption, most prominently on longer timescales (Figs. 3c and 5b). This effect can arise from overlapping contributions of GSB and PIA. In many reported SF materials, particularly rylene and DPP derivatives, the most intense triplet-state PIA overlaps with the GSB, and consequently, such spectra are typically regarded as a hallmark of triplet exciton formation. While sensitization experiments confirm that such spectral shapes can arise from triplets, we note that remarkably similar features can be generated by pump-induced transient thermal effects, which are insensitive to excitation power or repetition rate and thus behave like electronic population signatures in most crucial respects^38,42,43. One simple method to check their impact is to compare the long-lived TA spectrum and the first derivative of the absorption spectrum⁴². As can be seen in Figs. 3c and 4b, the line shapes of the two spectra are well-matched with each other for both film 1 and film 2, suggesting a significant contribution of thermal artifacts to the TA signals in the ground-state absorption region. The presence of these signals even at very low pump fluence, and their persistence throughout the ns-μs regime^38,39,42, significantly complicates any quantitative analysis of our data, for instance to calculate the triplet yield based on the singlet depletion method. To avoid any misinterpretation, we solely analyze the PIA band in the NIR region which is not overlapped with the GSB. Even here, caution is necessary: thermal signals can pull up NIR PIA bands towards positive ΔT/T, due to samples reflectance changes induced by the thermal change of refractive index^42,62,63. This case is prominent in film 2 (Fig. 5b), thanks to the absence of any long-lived PIA in the NIR.

One vs. two emissive states

We gain further insight into the triplet formation process from time-resolved PL spectroscopy on the ns timescale, which is not affected by such thermal artifacts and provides greater sensitivity over a wider dynamic range than TGPL³⁸. Figure 5 shows time-resolved PL results for films 1 and 2 obtained after 2.34 eV excitation with low pump fluence (2 µJ/cm²). We used the same 200 fs pump pulses as in TA, but here the time resolution is limited by the electronic gate of our ICCD detector to approximately 1 ns (a flat-top like IRF from −500 to 500 ps, Supplementary Fig. 13). In the case of J-like film 1, intense vibronic PL akin to the steady-state spectrum appears at time zero. Beyond the IRF, the signal magnitude is reduced by ∼99% similar to what we observed in TGPL, with weak delayed emission persisting up to 10 ns (Fig. 6a and Supplementary Fig. 14). Surprisingly, the spectral shape of the weak delayed PL strongly differs from both the prompt PL and the steady-state PL, showing a reversed vibronic peak ratio (Fig. 6c). Analyzing the spectral slices with area normalization, we can observe a clear isoemissive point near 1.7 eV (Fig. 6e). This interesting behavior suggests that there are two distinct emissive states contributing to the time-resolved PL kinetics⁶⁴. Moreover, we recall that the only electronic signatures detected in TA on these timescales are those for triplet excitons. This behavior indicates that the presence of triplets results in a distinct radiative decay pathway. In contrast to film 1, film 2 does not show any notable spectral evolution, namely the spectral shape of early and later delay times is nearly the same (Fig. 6b, d). Accordingly, the area-normalized analysis shows no isoemissive points, confirming that there is only one emissive state in film 2 (Fig. 6f).

Time-resolved PL results for a, c, e film 1 and b, d, f film 2 after photoexcitation of 2.34 eV at 2 µJ/cm². a, b Two-dimensional contour maps of the time-resolved PL data. c, d Comparison of intensity-normalized time-resolved PL spectra at the specified delay times. Because of low signal intensities, the spectra at 1.6 ns were measured in a separate experiment to achieve a sufficient signal-to-noise ratio. e, f Area-normalized time-resolved PL spectra from 0 to 2 ns with 0.1 ns interval. The dashed line in the left bottom panel for film 1 indicates an isoemissive point near 1.7 eV, supporting the presence of two emissive states.

¹(TT) emission

To understand the origin of the two emissive species in film 1, we repeat the experiment at a range of temperatures. Consistent with their assignment to different electronic states, the prompt and the delayed PL show distinct responses. Upon cooling, the prompt PL reveals an enhanced 0-0 peak relative to 0-1 vibronic emission (Fig. 7a). This behavior is in good agreement with the temperature-dependent optical response of J-aggregates⁶⁵, and it can be related to superradiance owing to reduced thermal disorder at low temperature. It provides an unambiguous demonstration that the prompt emission arises from the photoexcited lower Frenkel state. In the delayed PL, decreasing the temperature actually suppresses emission at the 0-0 peak and the lower vibronic progressions dominate the spectra (Fig. 7b). Together with the presence of only triplet-like PIA, this behavior is in excellent agreement with the previous reports of temperature-dependent ¹(TT) emission in (hetero)acenes, zethrenes, and conjugated polymers^15,16,66,67. In principle, emission from the ¹(TT) state to the ground state is symmetry-forbidden. However, the transition can become partially dipole-permitted by borrowing oscillator strength from a nearby state through Herzberg-Teller vibronic coupling, with the aid of non-totally symmetric vibrations^68,69. It has been known that this mechanism is also activated in carotenoids⁷⁰, cyclic conjugated oligomers^47,71, and some polycyclic aromatic hydrocarbons⁷², in which their S₁ states commonly have one-photon-forbidden nature. In this framework, the strong temperature dependence at the 0-0 peak is a signature of thermally activated repopulation of emissive lower Frenkel, and it suggests a small activation barrier of 33 meV (Supplementary Fig. 16)^15,16. Alternatively, following previous methods we can add one vibrational quantum to the 0-1 peak position of the ¹(TT) emission to estimate the ¹(TT) energy as 1.83 eV, compared to a singlet lower Frenkel energy (from PL) of 1.85 eV (Supplementary Fig. 16)¹⁶. Such weakly exothermic SF, similar to the case of F₂-TES-ADT (an anthradithiophene derivative), tetracene, TIPS-tetracene, and rubrene, helps to understand the heterogeneity of SF rates seen in Figs. 3 and 4, and the coexistence of singlet and triplet features over moderate time delays.

Temperature dependence of time-resolved PL in a, b, c film 1 and d, e, f film 2 after photoexcitation of 2.34 eV at 2 µJ/cm². The data were collected from 77 K to 297 K with 20 K interval. a, d and b, e representative normalized time-resolved PL spectra at early and later delay times, respectively. For film 1, prompt PL spectra are taken from 0 ns, and delayed PL spectra were obtained by averaging all spectra showing peak intensity <10⁻² in the normalized data. The black arrow in the middle panel for film 1 highlights the growth of the 0-0 band of the repopulated lower Frenkel emission via triplet-triplet annihilation (TTA) upon rising temperature. Film 2 does not show the distinct prompt/delayed behaviors so early and later spectra were taken from 0 and 1.6  ns, respectively. Panels c and f indicate integrated PL kinetics.

Despite the suitability of DPP dyes for SF and repeated study with TA spectroscopy^32,33, ¹(TT) emission has never been reported in this family of materials. This first and direct observation of ¹(TT) emission in a DPP thin film allows us to conclude that the triplet-like PIA observed in the TA data stems from SF, not from SO-ISC. This observation highlights the importance of directly correlating time-resolved PL with TA to understand SF and especially triplet-pair formation. Often the triplet features in TA are not prominent or are obscured by thermal effects, and very frequently the spectral differences between bound (i.e., ¹(TT)) and free triplets are too slight for reliable determination. Symmetry-forbidden emission in these instances provides a unique, often background-free signature of the ¹(TT) state.

Crucially, the delayed ¹(TT) emission decays at room temperature with a time constant of 1.2 ns, which is quantitively in line with the TA kinetics in the ¹(TT) PIA band at the NIR region from 1 to 7.6 ns (Fig. 6c and Supplementary Fig. 17). On lowering the temperature, the lifetime of ¹(TT) emission is enhanced, presumably due to suppression of other non-radiative deactivation channels at low temperature, but we still observe essentially complete decay within 50 ns. For comparison, we note that there are no significant temperature-dependent spectral changes in film 2 (Fig. 6b). We only detect slight bandwidth changes, implying that energy distribution on the potential energy surface of the excimer-like state is modulated controlled by temperature. As in film 1, cooling increases the delayed emission lifetime, which we again ascribe to a reduction in non-radiative deactivation rates.

Long-lived triplets born through SO-ISC

To learn more about the fate of ¹(TT) we carry out trEPR spectroscopy on film 1 and, for comparison, film 2. We focus our analysis of film 1 on the spectrum acquired at 50 K with excitation at 1.91 eV (Fig. 8a), though we observe negligible excitation energy or temperature dependence. Our measurements show a weak but clear triplet state spectrum with a polarization pattern of eea/eaa (where e is emission and a is enhanced absorption). This spectrum can be described by zero-field splitting parameters (ZFS) of D ≈ 1150–1250 MHz and |E| ≈ 290–360 MHz, with the uncertainty dominated by the low signal level and partial ordering effects. Our E value is in good agreement with previously determined ZFS parameters for the triplet in a related TDPP derivative, formed via charge recombination in a blend film with PC₇₀BM. The reported D = 1550 MHz is distinctly higher⁵⁹, suggesting that the extended side chains in HR-TDPP-TEG permit greater delocalization of the spin density. Crucially, the observed polarization pattern of eea/eaa indicates the mechanism for triplet formation in film 1 is SO-ISC. The inner canonical peaks are not apparent when excitation is at 2.33 eV, suggesting a slightly different zero-field triplet sublevel population. The polarization pattern whether eea/eaa or eee/aaa corresponds to an SO-ISC-born triplet, that is, the triplets observed on these timescales do not arise from SF. This can be seen unambiguously by comparison to model spectra for the two pathways (Fig. 7b), where SF-born free triplets exhibit a unique polarization pattern due to direct population of the high-field eigenstates (aee/aae for D > 0)^18,19. See Supplementary Note 6 for further details on the model spectra.

a Averaged trEPR spectra (0.1–2.8 µs) of film 1 and film 2. Temperature and applied excitation energy are indicated below each spectrum. All spectra were measured with an angle of 0° relative to the external magnetic field direction. See Supplementary Information for further details. b Representative simulation of SO-ISC- and SF-born triplets using EasySpin with ZFS parameters \(D=1200\) MHz and \(E=-320\) MHz. The blue and red lines represent the simulated spectra of the T₋ ↔ T₀ and T₊ ↔ T₀ components, clearly showing the polarization pattern.

To extend our search for SF signatures, we repeated these measurements at different temperatures. At 20 K, in addition to increased triplet lifetime and spin-relaxation times there is precedent to observe long-lived ⁵(TT). This state, unique to SF systems, yields additional features with a spectral width of D/3. Yet we observe no significant triplet spectral changes at 20 K (Supplementary Figs. 22 and 23). At 290 K the conditions approximate those used for time-resolved optical spectroscopy, and again we detect no difference in the extremely weak triplet polarization pattern (Supplementary Figs. 24 and 25). As in the optical experiments, we find that excitation energy has no effect. Strikingly, we even observe similar triplet features—polarization pattern and spectral width—in H-like film 2, which shows no evidence for SF and no detectable optical signatures of triplet formation. Because of this surprising behavior, we additionally considered the role of orientation of both samples relative to the external magnetic field, which has been shown to impact the trEPR spectra of triplets in other SF materials⁷³. In neither film did we observe the appearance of SF spectral signatures at other angles. Full details on the trEPR measurements at different temperatures, rotation angles, and pump energies are provided in the Supplementary Information. From this lack of dependence on temperature, excitation energy, orientation, or even intermolecular coupling motif, we can unambiguously conclude that the evolution of ¹(TT) into free triplets is not a preferred pathway in film 1. In both films, the very small population of triplets must arise from SO-ISC. Combining with ns-TA results, although it is not possible to precisely quantify the triplet yield due to the thermal artifacts, we highlight that the yield from the SO-ISC pathway is likewise evidently very small (Supplementary Fig. 12d). Furthermore, a CT state signal is present in both samples at low temperatures (See Supplementary Note 7).