Transferring OLED Roll-Off Physics to Predict Low-Energy Quenching in Organic Scintillators

When a light dark matter particle hits a heavy nucleus such as xenon or germanium, it hands over almost no energy, like a ping-pong ball bouncing off a bowling ball. Hydrogen-rich organic scintillators fix the kinematics. A proton is closer in mass to a sub-GeV particle, so it recoils with far more energy. The catch is that slow, dense proton tracks lose most of their light to quenching, and the scale of that loss at sub-keV energies is the dominant uncertainty. Organic-LED engineers study the same underlying process under another name. In an OLED, efficiency roll-off is the loss of brightness at high current, and it comes largely from excited molecular states (excitons) colliding and destroying each other: triplet–triplet and singlet–triplet annihilation. The display community has learned to measure these collision rates cleanly and to design molecules that suppress them.

This note proposes a transfer test. First, measure annihilation rate constants in electrically and optically driven thin films of candidate emitters. Second, compute the initial exciton density along a recoil track from first-principles electronic structure, meaning density functional theory (DFT), a method that computes a material's electrons from its electron density. Third, combine the two with no fitted parameters to predict the scintillation quenching curve, then check that prediction against ion-beam measurements. If the transfer works, the design rules behind low-roll-off OLED emitters become design rules for low-threshold organic targets. The central molecular hypothesis is that emitters built for fast triplet-to-singlet conversion should also quench less under low-energy recoils.

Core Claim

Scintillation quenching in organic films is, to first order, the same bimolecular annihilation process that causes OLED efficiency roll-off. The only difference is the geometry of the exciton population: a dense cylinder around a track instead of a uniform layer. Two consequences follow if this is right.

The first is predictive. Annihilation constants extracted from roll-off curves, combined with a DFT-computed initial track profile, should predict relative light yield versus recoil energy without fitting any quenching parameter to scintillation data. The empirical alternative, Birks' law, fits its constant material by material.

The second is a design consequence. Hybridized local and charge-transfer (HLCT) emitters, which combine a bright localized excitation with a charge-transfer character that speeds conversion of triplets into singlets, should retain more light at low recoil energy than conventional aromatic scintillator molecules and doped delayed-fluorescence emitters. Xu reports such a molecule, the non-doped emitter TPAAnPI. Molecular DFT on TPAAnPI identified a high-lying triplet-to-singlet channel ($T_{12}\rightarrow S_3$) that yields 91% exciton utilization, and the device shows only 15% roll-off at 1000 cd/m².

The transfer is physically plausible because both excitation routes produce roughly 25% singlets and 75% triplets. OLED excitons form from electron–hole recombination, and ionization tracks also produce excitons largely through recombination of freed charges.

Relation to Prior Work

The literature supplying the recoil-side method is the dielectric program of Knapen, Kozaczuk and Lin. Migdal Effect in Semiconductors expresses Migdal electron excitation in silicon and germanium through the energy loss function (ELF), $\text{Im}[-1/\epsilon(\omega,\mathbf{k})]$, a quantity describing how readily the material absorbs a given energy and momentum. Dark matter–electron scattering in dielectrics shows that in-medium screening suppresses rates by a factor of 1.4–2.5 at 10 MeV, with dielectric functions computed from first principles in GPAW. DarkELF packages these calculations for eight inorganic targets. We reuse the ELF machinery for a different purpose: not event rates, but where and how densely the deposited energy lands. All three papers stop once charge carriers or phonons are produced, and none treats an organic target or light output. Liang et al. constrains the scope. They found that free-recoil and phonon-mediated Migdal rates agree in silicon down to 10 MeV, but only for a monatomic crystal at zero temperature. We therefore do not claim to settle Migdal physics in molecular crystals; we treat nuclear-recoil light yield as the primary observable.

The motivating evidence for a quenching gap comes from direct-detection calibration. Huang et al. find that the NEST simulation toolkit tracks PandaX-II light yield across electric fields, from 69.5 to 55.6 photons/keV at 41.5 keV, within 4–8% uncertainty. Their 9.4 keV point, however, shows recombination enhancement the model does not predict. Dutta et al. rely on an externally measured, energy-dependent quenching factor for COHERENT CsI, which is one example of the field treating quenching as an empirical input.

The roll-off literature supplies the rate physics. Ihn et al. and Li et al. attribute roll-off to triplet–triplet and singlet–triplet annihilation driven by long-lived delayed-fluorescence states. Zhou et al. and Li et al. report near-zero roll-off in non-doped HLCT emitters with balanced carrier transport. None of this work has been connected to ionizing tracks.

Research Program

The model has three stages, and every parameter comes from an independent measurement or calculation.

The first stage is the initial track. For a recoiling proton or carbon ion of energy $E$, we need the energy split between electronic excitation and nuclear (heat) losses, plus the radial profile of the resulting excitations. The electronic stopping and the spectrum of secondary excitations follow from the ELF integrated over momentum transfer. For the organic films we compute the ELF with GPAW time-dependent DFT, extending DarkELF's grid, and cross-check it against the Mermin model fitted to measured optical constants. The known band-gap underestimate of standard exchange-correlation functionals is corrected with a scissor shift pinned to the measured optical gap.

The second stage is annihilation kinetics. Singlet density $n_S$ and triplet density $n_T$ evolve in a cylindrical track according to

$$\partial_t n_T = D_T\nabla^2 n_T - \frac{n_T}{\tau_T} - k_{RISC}\,n_T - k_{TT}\,n_T^2 - k_{ST}\,n_S n_T ,$$

together with a matching singlet equation carrying radiative rate $1/\tau_S$. Here $D_T$ is the triplet diffusion constant, $\tau_T$ the triplet lifetime, $k_{RISC}$ the reverse intersystem-crossing rate, and $k_{TT}$ and $k_{ST}$ the annihilation constants. Relative light yield is the integrated singlet radiative emission divided by the same quantity at vanishing density.

The third stage is sourcing the constants. $k_{TT}$, $k_{ST}$, $\tau_T$ and $k_{RISC}$ come from fits to optically pumped roll-off curves, meaning photoluminescence efficiency versus pump density, and from transient photoluminescence on films of the same morphology. Electrically driven OLED roll-off serves as a consistency check. Optical pumping is primary because electrical roll-off also contains exciton–polaron quenching and charge imbalance, which would contaminate the transfer. Molecular DFT supplies singlet and triplet energies and the high-lying conversion channels, and it is used to shortlist emitters, not to fit anything.

The material set spans the hypothesis. It includes TPAAnPI and a TPACFTAZ-type HLCT emitter, a doped thermally activated delayed-fluorescence film with millisecond-scale triplets (predicted worst), and a conventional aromatic scintillator film (the incumbent). All are evaporated as micron-thick films so that ion tracks stop inside the film.

Baselines

The naive reference is a linear response: light proportional to deposited electronic energy with no quenching, meaning the quenching factor is set only by the electronic-to-nuclear split. Any credible model must beat it at every energy, and it tells us how much of the quenching is kinetic rather than kinematic.

The strongest existing competitor is Birks' law with its constant $k_B$ fitted to each material's own ion-beam data. This is how scintillator response is actually modeled, and it encodes bimolecular quenching phenomenologically. For fairness, Birks gets the same stopping-power input as our model and its constant is fitted on a training energy range, typically above 100 keV. Both models are then scored on held-out low energies. The proposal beats Birks if our zero-free-parameter prediction has lower held-out error in relative light yield in the lowest energy decade, or if a single $k_B$ cannot fit the delayed-fluorescence and HLCT films while our model tracks both. We found no reported result on Lacuna for Birks-law accuracy on organic proton-recoil quenching, so this comparison is measured, not quoted. For noble liquids the nearest reported reference is NEST's 4–8% light-yield agreement in PandaX-II from Huang et al.. That is a different material class and serves only as context for how well a mature empirical model can do.

Two ablations isolate the key ingredients. The first replaces the DFT energy loss function with the Mermin fit and with a plain tabulated stopping power, which tests whether first-principles track structure matters at all. The second sets the triplet channels to zero, dropping $k_{TT}$ and $k_{ST}$ and keeping only singlet–singlet annihilation, which tests whether the OLED-specific triplet physics is what carries the prediction.

Why not just…?

Why not just fit Birks' law to each new material? For a material already synthesized and calibrated, that works, and it is the benchmark above. It does not rank molecules before they are made. It also cannot distinguish two emitters with identical stopping power but different triplet lifetimes, which is exactly the difference the design question depends on. If our held-out comparison shows Birks matching our accuracy and ordering the materials correctly after the fact, the predictive claim reduces to "roll-off data rank scintillators." That weaker claim would still be useful, and we would report it as such.

Why not just measure quenching factors directly with neutron or ion beams? Direct measurement is the ground truth and is our test data. It is expensive per material, and it becomes statistically weak exactly in the sub-keV region that matters for light dark matter. The model's value is screening candidates from cheap tabletop roll-off measurements and extrapolating below measurable energies. We check that extrapolation by holding out the lowest measured energies.

Why not just use tabulated stopping powers and skip DFT for the track? This is a real threat. Track radius may be dominated by secondary electrons whose range depends weakly on band-structure detail. The first ablation settles the question. If tabulated stopping matches the DFT energy loss function within measurement error, we drop it and DFT's role shrinks to molecular screening of triplet channels, which is the honest outcome.

Proposed Experiments

The first experiment tests transfer. For each film we measure optically pumped roll-off and transient photoluminescence, then predict relative light yield versus energy for protons and helium ions. Those predictions are compared against ion-beam light-yield measurements on the same films, using a held-out low-energy range. The decisive outcome is that the predicted rank order of quenching across the four materials matches measurement, and that the predicted absolute curve falls within the combined uncertainty with no scintillation-fitted parameter.

The second experiment tests the design hypothesis directly. If fast triplet conversion is what suppresses quenching, the HLCT films should show the highest low-energy relative yield and the doped delayed-fluorescence film the lowest. The model's $k_{RISC}$ sensitivity should explain the gap. The hypothesis is falsified if the HLCT films are no better than the aromatic reference.

The third experiment checks consistency between excitation routes. Electrically extracted constants should agree with optically extracted ones after the polaron term is accounted for. If they disagree by more than a factor of two, electrical roll-off is not a usable proxy, and the program rests on optical pumping alone.