Detectors that hunt light dark matter have to see nuclear recoils of tens to hundreds of electron-volts: the tiny kick a nucleus gets when a light particle bounces off it. Nobody has measured cleanly how much of that energy shows up as a readable signal at these scales. The obvious way to measure it is to aim a beam of ions of one known energy at the detector crystal. A popular version of the idea says that laser-plasma accelerators, which drive ions with extremely strong fields made when a laser hits a thin foil, could supply such beams compactly. This proposal argues that the popular version is aimed at the wrong target, and that a better one exists.
The popular version fails for two reasons. A charged ion fired into a crystal loses most of its energy directly to electrons along its path, and that signal swamps the effect one wants to calibrate. And conventional electrostatic accelerators already make ions with far better energy purity than any laser source. The better target is to use the laser for the one thing it does well: emitting all of its protons within picoseconds. Those protons strike a lithium layer just above the threshold of the 7Li(p,n) reaction, which produces a flash of forward-going neutrons of a few to about a hundred keV. Because the flash is so short, each neutron's energy can be read off its flight time over tens of centimeters, and the recoil energy it gives a silicon or germanium nucleus can be fixed by detecting the scattered neutron at a known angle. The broad energy spread of the laser protons, the property most papers try to suppress, then does not have to be removed. It is read out shot by shot through timing. Particle-in-cell simulation, a plasma model that tracks many representative charged particles moving in self-consistent electromagnetic fields, supplies the source description that this whole chain depends on.
Core Claim
A laser-driven proton source feeding a near-threshold lithium converter can produce time-tagged neutrons with enough per-event energy resolution to calibrate sub-keV nuclear recoils in germanium and silicon. The resolution comes from flight time, not from making the protons monoenergetic. The claim is only worth making if this source beats a conventional pulsed accelerator on one figure of merit: tagged recoils per hour at a fixed recoil-energy resolution, per unit of floor space. The program is built so that a negative answer is clear and publishable.
The kinematics show why sub-keV recoils are reachable. A neutron of energy $E_n$ scattering elastically by lab angle $\theta$ off a nucleus of mass $m_N$ transfers approximately
$$E_{nr} \approx \frac{2 m_n m_N}{(m_n+m_N)^2}\,E_n\,(1-\cos\theta),$$
where $m_n$ is the neutron mass. For a neutron near 30 keV, which is roughly where the 7Li(p,n) reaction emits just above threshold, this gives a maximum of about 4 keV in silicon and about 1.6 keV in germanium. Forward-angle tagging reaches tens of eV. The neutron energy comes from $E_n = \tfrac{1}{2} m_n (L/t)^2$ over flight length $L$ and flight time $t$, so the fractional energy error is roughly twice the fractional timing error. A neutron of a few tens of keV travels at a few million meters per second, so it needs a few hundred nanoseconds to cross half a meter. A picosecond-scale emission time therefore leaves the resolution limited by the detector's own timing rather than by the source. That same delay of hundreds of nanoseconds after the laser shot also separates the neutron's arrival from the prompt electromagnetic pulse of the laser shot.
Why This Is Different From the Obvious Idea
The research file points toward the "three-way intersection" of monoenergetic laser-driven ion beams, modeled by particle-in-cell codes, used to calibrate Migdal thresholds. The Migdal effect is the extra electronic excitation that happens when a struck nucleus jerks away from its own electron cloud. This proposal departs from that intersection in two deliberate ways.
First, it does not try to make low-energy monoenergetic ions. The work on energy-spread suppression is impressive. Shock acceleration in tailored plasma profiles reaches about 10% FWHM in Laser-Driven Shock Acceleration of Monoenergetic Ion Beams, and sharp-front circularly polarized pulses stabilize light-sail acceleration in Efficient acceleration of monoenergetic proton beam by sharp front laser pulse. Both methods, however, target energies from tens of MeV to GeV, and both depend on idealized targets. A 10% spread at keV energies would still lose badly to an electrostatic machine.
Second, it does not fire charged ions at the detector. The solid-state Migdal theory is the reason this matters. That theory expresses the effect through the crystal's energy loss function, meaning how readily the crystal's electrons absorb a given energy and momentum; this is the approach of Migdal Effect in Semiconductors, implemented in DarkELF. A related line recasts the effect as bremsstrahlung-like emission with an $\omega^{-4}$ spectrum, in Describing the Migdal effect with a bremsstrahlung-like process and many-body effects and Phonon-mediated Migdal effect in semiconductor detectors. All of these describe a neutral projectile suddenly kicking a nucleus. A keV ion inside the lattice excites electrons along its own path through the same dielectric response, with no suppression factor, so any Migdal signal would be buried. Neutrons avoid this problem, and they are also what the dark-matter analyses themselves assume.
The citations play distinct roles. The acceleration papers supply the source physics, and the fact that they have not reached keV motivates the timing approach. The DarkELF and Liang papers supply the predictions to be tested. Calibration of CR-39 solid state track detectors with monoenergetic protons from 0.3 MeV to 2.5 MeV constrains the design twice. It reports that electronic sensors fail in laser chambers from electromagnetic pulses, and it provides the passive track-detector calibration we will use to measure the proton spectrum at exactly the 1.9 MeV range that matters.
Research Program
The first stage is source modeling. We will run 2D particle-in-cell simulations, with selected 3D confirmation, in an OSIRIS-class code of the kind used by Fiuza et al. The simulated setup is rear-surface sheath acceleration from micron-scale foils at normalized laser amplitudes of a few. For readers new to the field: the laser heats electrons, which escape the back of the foil and set up a charge-separation field that pulls protons after them. These simulations do not aim to reach high energy. They must predict four quantities: the number of protons per shot in the 1.88–1.95 MeV window, the emission duration and divergence of those protons, the co-moving hot electrons that reach the converter, and the timing and spectrum of escaping electrons that seed the electromagnetic pulse.
The second stage is transport and conversion. The simulated proton phase space is passed to a Geant4 model of a thin LiF or lithium converter a few millimeters to centimeters downstream. This model gives the neutron yield, angle, and emission time. It also accounts for protons above the collimation window, which make faster neutrons that timing will sort out rather than discard.
The third stage is the detector forecast. A germanium point-contact detector and a silicon charge-readout detector are modeled at 30–50 cm, with an array of scattered-neutron scintillators at chosen angles. The recoil spectrum is folded through two models. One is the quenching assumed in current analyses. The other adds the Migdal ionization predicted by DarkELF and by the $\omega^{-4}$ bremsstrahlung picture. The output is the number of shots needed to separate these predictions at a given recoil energy.
The fourth stage is contingent on the first three passing their go/no-go thresholds. It is a staged laboratory measurement of the proton spectrum with calibrated CR-39, then the neutron time-of-flight spectrum with scintillators, then the germanium detector behind shielding and a gated readout.
Baselines
The naive reference point is what large experiments do now. They calibrate with broadband neutron sources: AmBe, deuterium–deuterium (DD) generators, and 252Cf. PandaX-4T used AmBe and DD sources for its nuclear-recoil band and background modeling, reporting a neutron-induced background of 1.15 ± 0.57 events over 5–100 keV of nuclear recoil energy in Neutron-induced nuclear recoil background in the PandaX-4T experiment. Its light-dark-matter search then reached nuclear recoils of about 0.33–3.0 keV in Search for Light Dark Matter with 259 Days of Data in PandaX-4T. These sources give no per-event recoil energy. We will model them in the same Geant4 geometry and compare per-event recoil-energy uncertainty against the tagged laser source.
The strongest competitor is a conventional pulsed electrostatic accelerator driving the same near-threshold 7Li(p,n) converter, with nanosecond beam pulses and the same scattered-neutron tagging. It is strongest because its energy spread, pulse repetition, and electromagnetic quiet are all better than a laser's, and only compactness and emission duration favor the laser. For a fair comparison, both sources feed the identical converter, flight path, detector, and tagging array in simulation. Each source's free parameters (flight length, converter thickness, and laser amplitude or beam current) are tuned with the same optimization budget. The figure of merit is tagged recoils per hour below 500 eV at 10% recoil-energy resolution. The file contains only electrostatic proton calibration, not keV neutron tagging. The tandem calibration of CR-39 used 12 energy points from 0.3 to 2.5 MeV at about 5×10⁴ protons per detector, per He et al. There is no reported result found on Lacuna for pulsed-accelerator tagged keV-neutron recoil calibration, so this baseline's rate will be built from standard beam currents in our own model and stated as such. The laser source beats it only if it reaches at least equal figure of merit at an achievable repetition rate, or matches it within a factor of a few at a much shorter flight path. Otherwise the conclusion is that compact co-location is not worth the cost.
Two ablations isolate the key ingredient. In the first, the laser source runs without time-of-flight: neutron energy is inferred only from the simulated proton spectrum. This shows how much of the resolution comes from timing. In the second, the plain sheath-acceleration source is replaced by a spread-suppressed shock-acceleration source of the Fiuza type, scaled down in energy. This tests the central thesis that suppressing spread buys little once timing is used. If the suppressed source gives a large gain, the thesis is wrong and the energy-spread direction becomes central again.
Why not just…?
Why not just use a conventional electrostatic accelerator? This is the real threat, and the proposal is built around it. Electrostatic machines give sub-percent energy spread, quiet electronics, and high duty cycle. The laser's only advantages are picosecond emission and compactness, which matters for underground sites such as CJPL, home to CDEX-50 in Projected WIMP sensitivity of the CDEX-50 dark matter experiment. The matched-geometry figure-of-merit comparison above settles it. If the pulsed accelerator wins by more than an order of magnitude, we will report that directly.
Why not just fire keV ions straight into the crystal to see the Migdal effect? The projectile's own electronic stopping excites the same electron–hole pairs that the Migdal effect would produce, through the same energy loss function computed in DarkELF. This contribution has no suppression and would dominate. The forecast stage will quantify this by running the charged-ion case through the same dielectric model and showing the size of the signal it would bury.
Why not just keep using AmBe or DD neutron sources? Their neutron spectra are continuous and the scattering angle is unknown for each event, so the recoil energy is known only statistically. Multiple scattering in the target, the "multi-scatter" category PandaX-4T had to subtract, further blurs the sub-keV region. The naive-baseline comparison will measure this as per-event recoil-energy uncertainty.
Why not just use an analytic sheath-expansion model instead of particle-in-cell? For the proton spectrum alone, an analytic model might be good enough, and we will test that by comparing its predicted counts in the 1.88–1.95 MeV window to the simulation's. Particle-in-cell is needed for what the analytic model does not give: the co-moving electron population and escape timing that determine electromagnetic-pulse gating, and the divergence at the converter. If the analytic model reproduces all figure-of-merit inputs to within 20%, we will adopt it for the design sweeps and keep particle-in-cell for spot checks.
Proposed Experiments and Decisive Checks
We first validate the simulation code by reproducing the roughly 10% FWHM spectrum and 2–3% conversion efficiency reported by Fiuza et al. in their shock-acceleration regime. Only then do we trust it in the sheath regime.
The first decisive simulation check asks whether protons per shot in the near-threshold window, times the converter yield, times the tagging acceptance, gives enough tagged sub-500 eV germanium recoils. The requirement is enough to separate the DarkELF Migdal prediction from a no-Migdal hypothesis at two to three standard deviations within a feasible number of shots.
The second check asks whether the electromagnetic pulse, modeled from the escaping electron population, has decayed below the germanium readout's recovery level before the earliest neutron arrival of interest. The paths are shielded and the readout is gated.
In the laboratory stage, the proton spectrum is measured with CR-39 under He et al.'s etching protocol, within the 0.3–2.5 MeV range they calibrated. This test fails if the measured window population falls more than a factor of three below the simulation. The neutron time-of-flight spectrum must then show the kinematically collimated keV component. Only after both pass does the germanium detector go on the line. Its first physics target is the ionization yield per tagged recoil from 50 to 500 eV, compared against quenching-only and Migdal-augmented predictions, including the $\omega^{-4}$ low-energy shape of Liang et al. An outcome where the laser source reaches the tagged rate but loses to the pulsed-accelerator baseline still gives the field a quantitative answer on whether compact laser sources belong underground.
References
- Laser-Driven Shock Acceleration of Monoenergetic Ion Beams
- Efficient acceleration of monoenergetic proton beam by sharp front laser pulse
- Migdal Effect in Semiconductors
- python package for dark matter scattering in dielectric targets
- Describing the Migdal effect with a bremsstrahlung-like process and many-body effects
- Phonon-mediated Migdal effect in semiconductor detectors
- Calibration of CR-39 solid state track detectors with monoenergetic protons from 0.3 MeV to 2.5 MeV
- Neutron-induced nuclear recoil background in the PandaX-4T experiment*
- Search for Light Dark Matter with 259 Days of Data in PandaX-4T
- Projected WIMP sensitivity of the CDEX-50 dark matter experiment