If dark matter is lighter than a proton, its best chance of being seen is to knock an electron loose inside a crystal. The predicted rate of such events depends on how the crystal's electrons respond collectively to a kick of a given energy and momentum. Physicists summarize that response in one function, the energy loss function (ELF). Today the ELF is computed, not measured, at the momenta that matter. Different computational recipes disagree, and those disagreements become uncertainty in every exclusion limit. Angle-resolved photoemission spectroscopy (ARPES), a technique that shines light on a crystal and records the energy and exit angle of the electrons it ejects, directly maps where electrons actually sit in energy and momentum, and how sharply defined those states are. We propose to use ARPES maps to fix the measurable ingredients of the ELF calculation. We then test, on data the calibration never saw, whether this shrinks the theoretical uncertainty on sub-GeV dark-matter rates.
In two sentences: the rate predictions for light dark matter in silicon and germanium rest on computed electron band structures and a hand-tuned broadening parameter, and nobody has checked those inputs against measured band maps. We correct the inputs with ARPES, recompute the response with its full directional dependence, and ask whether the corrected prediction agrees better with independent optical data and how far it moves the predicted dark-matter reach and daily modulation.
Core Claim
The dark-matter literature has already shown that the crystal's response is not a detail. Dark matter-electron scattering in dielectrics recasts electron scattering in terms of the ELF, $\mathrm{Im}[-1/\epsilon(\omega,\mathbf{k})]$, where $\epsilon$ is the dielectric function at energy transfer $\omega$ and momentum transfer $\mathbf{k}$. It finds that screening suppresses rates by a factor of 1.4 to 2.5 for 10 MeV dark matter in Si and Ge. It computes the ELF two ways: density functional theory (DFT, the standard first-principles band calculation) in the random phase approximation, and a Mermin oscillator model fitted to optical data. The two agree poorly at high energy and momentum, and the response is averaged to be isotropic. In-medium screening effects for the Galactic halo and solar-reflected dark matter detection in semiconductor targets adds local field effects, which change rates "within a factor of 0.5". It reports that low-mass limits depend on an unverified broadening parameter $\eta$. DarkELF packages these recipes for eight targets and shows the same divergences. These three papers motivate our proposal: they locate the uncertainty in single-particle inputs (band positions, band widths, lifetimes) that no one has compared with measurement.
Our claim is narrower than "ARPES measures the ELF," which it does not. ARPES measures the occupied spectral function: band energies $E(\mathbf{k})$ and linewidths across the Brillouin zone. The ELF is a two-particle quantity that also needs the empty conduction bands. The proposal is that ARPES fixes three things the dark-matter calculations currently guess:
- Valence-band dispersion over the full zone. The relevant momentum transfers, of order $\alpha m_e \approx 4$ keV, span a large fraction of the zone, so the whole valence band matters.
- Energy-dependent quasiparticle broadening. This replaces the constant $\eta$.
- Band-resolved energy renormalization.
The conduction bands are anchored separately by the optical gap. The result is a direction-resolved ELF whose remaining freedom is small and stated explicitly.
Research Program
We start with Si and Ge, because they are the targets of the cited calculations and are precomputed in DarkELF. Bulk-sensitive soft X-ray ARPES, with photon-energy sweeps, recovers the momentum perpendicular to the surface ($k_z$). The usual high-resolution vacuum-ultraviolet ARPES probes only the top 5–10 Å and would mix in surface states. From these maps we fit a low-dimensional correction to a DFT band structure. The correction consists of band-dependent stretch factors and energy offsets, plus a broadening function $\eta(E)$ read from measured linewidths after removing instrumental resolution. Matrix-element effects, meaning intensity changes with light polarization and photon energy, can fake gaps. We therefore fit only peak positions and widths, never intensities, and require consistency across two polarizations.
The corrected bands feed a random-phase-approximation ELF calculation with local field effects included. We keep the full $\mathbf{k}$ dependence instead of the isotropic average. We then propagate the ELF through the DarkELF rate machinery for electron scattering and the Migdal channel, the process in which a nuclear recoil shakes loose an electron. For the Migdal channel we restrict to masses above about 70 MeV, where Migdal Effect in Semiconductors shows its impulse approximation holds. Finally, we compute the directional rate as the Earth's velocity through the galactic halo rotates relative to the crystal axes. This gives a predicted daily modulation amplitude. QROCODILE highlights directional response as a background-rejection tool, but its prediction comes from sensor geometry, not band structure.
A second stage repeats the procedure on one strongly anisotropic layered crystal. In cubic Si the directional effect should be small, which is itself a prediction we check. The case for ARPES-anchored anisotropy is strongest where ARPES has already found Fermi-surface and band anisotropy that computation places poorly, as in the open-orbit semimetal MoAs2. We deliberately leave altermagnetic targets to later work, because spin-dependent dark-matter couplings would need new theory before ARPES could inform them.
Baselines
The naive reference point is the isotropic DFT random-phase ELF without local field effects and with a constant $\eta$, as shipped in DarkELF. Every improvement is measured against it.
The strongest computational competitor is a many-body (GW) quasiparticle calculation followed by the same random-phase ELF with local field effects. GW is a perturbative correction to DFT that fixes band gaps and band widths from first principles. It attacks exactly what ARPES corrects, with no experiment. We run it on the same crystal structures, the same k-point meshes and the same rate code. Only the source of the band correction differs.
The strongest empirical competitor is the Mermin oscillator fit to optical data from Dark matter-electron scattering in dielectrics. It is exact at zero momentum by construction, so we compare against it only at finite momentum and on observables it did not fit.
For the strongest published effect size the comparison point is the 1.4 to 2.5 screening suppression at 10 MeV and the factor-0.5 local-field variation. No reported result was found on Lacuna for the absolute error of any ELF method against measured finite-momentum loss spectra. Establishing that number is part of the work.
An ablation isolates the key ingredient. We run ARPES dispersion corrections with constant $\eta$, and constant bands with ARPES-derived $\eta(E)$, to see which input carries the change.
Beating the baselines means two things. First, the ARPES-anchored ELF must predict held-out optical-limit data (the zero-momentum ELF and the onset of absorption) with lower error than DFT and GW, without having been fitted to them. Second, the spread of predicted 10 MeV rates across reasonable calibration choices must be narrower than the current spread across the three existing recipes.
Why not just…?
Why not just use GW instead of measuring anything? This is the real threat. For Si and Ge, GW may reproduce ARPES dispersions within resolution. If the held-out comparison shows GW and the ARPES-anchored ELF within experimental error, we will report that ARPES is unnecessary for simple semiconductors. The effort then moves to the anisotropic and correlated targets. For such materials, ARPES work on YbV6Sn6 and CrSb shows that computed bands need experimental anchoring. GW also does not supply the lifetime broadening that replaces $\eta$. The ablation tests whether broadening alone matters.
Why not just fit Mermin oscillators to optical or electron-energy-loss data? Optical data live at zero momentum. The Mermin extension to finite momentum is a phenomenological extrapolation that diverges from DFT at large $\mathbf{k}$ and is isotropic by construction. It therefore cannot predict a daily modulation. Our held-out test reverses the usual roles: optical data become the referee rather than the input.
Why not just measure the ELF directly at finite momentum with inelastic X-ray or momentum-resolved electron-energy-loss scattering? These techniques measure the right quantity, and where such spectra exist they are the best possible validation target. We would use them as a second held-out test. They are typically taken along a few crystal directions with coarser energy resolution near the gap. ARPES-anchored bands instead give a calculation that can be evaluated in every direction. If direct loss spectra prove dense enough to build the directional rate on their own, that would be a better route than ours, and we would say so.
Why not just build bigger liquid-xenon detectors or denoise existing data? Xenon recoils fall below threshold for sub-GeV masses (PandaX-I). Denoising, as in TIDMAD, with a denoising score of 6.43 against 1.0 for no denoising, improves the signal-to-noise ratio but not the predicted signal against which a limit is set.
Proposed Experiments
The decisive check is the held-out prediction. ARPES calibration uses only photoemission peak positions and widths. The resulting ELF must then match measured optical absorption and the zero-momentum loss spectrum of Si and Ge better than the DFT and GW baselines. The metric is the rate-weighted fractional error over 1–50 eV, where the weight is the 10 MeV dark-matter rate kernel so that errors count where they move limits.
The second check propagates calibration uncertainties to predicted rates at 1, 10 and 100 MeV, for both a heavy and a light mediator. We report whether the resulting band is tighter than the spread among the Lindhard, Mermin and DFT recipes.
The third check is directional. We predict the daily modulation amplitude for Si, where a small value is expected, and for the anisotropic crystal. We also report the difference from the isotropic average.
The proposal fails cleanly if the ARPES-anchored ELF does no better than GW on held-out data in either material class.
References
- Dark matter-electron scattering in dielectrics
- In-medium screening effects for the Galactic halo and solar-reflected dark matter detection in semiconductor targets
- python package for dark matter scattering in dielectric targets
- Migdal Effect in Semiconductors
- First Sub-MeV Dark Matter Search with the QROCODILE Experiment Using Superconducting Nanowire Single-Photon Detectors
- Observation of open-orbit Fermi surface topology in the extremely large magnetoresistance semimetal<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:msub><mml:mi>MoAs</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:math>
- Strongly Entangled Kondo and Kagome Lattices and the Emergent Magnetic Ground State in Heavy-Fermion Kagome Metal <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:mrow><mml:mrow><mml:msub><mml:mrow><mml:mi>YbV</mml:mi></mml:mrow><mml:mrow><mml:mn>6</mml:mn></mml:mrow></mml:msub></mml:mrow><mml:mrow><mml:msub><mml:mrow><mml:mi>Sn</mml:mi></mml:mrow><mml:mrow><mml:mn>6</mml:mn></mml:mrow></mml:msub></mml:mrow></mml:mrow></mml:math>
- Large Band Splitting in <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:mrow><mml:mi>g</mml:mi></mml:mrow></mml:math>-Wave Altermagnet CrSb
- Low-mass dark matter search results from full exposure of the PandaX-I experiment
- TIDMAD: Time Series Dataset for Discovering Dark Matter with AI Denoising