Closed Dielectric Haloscope Research

• Closed dielectric haloscopes are devices that convert axion dark matter into microwaves using dielectric interfaces and reflective boundaries.
• They employ mirror-terminated multilayer stacks or metal-enclosed cavities to shape resonant modes, achieving high boost factors and quality factors for signal enhancement.
• Practical implementations such as MADMAX and QUAX–aγ illustrate the balance between bandwidth, alignment, and calibration crucial for improving dark matter detection sensitivity.

A closed dielectric haloscope is a haloscope architecture in which dielectric interfaces operate under a reflective or enclosing electromagnetic boundary so that axion-induced radiation is either directed into a controlled output port or stored in a cavity-like mode. In the literature, the term is applied both to mirror-terminated dielectric stacks used as one-sided boosters and to metal-enclosed dielectric-loaded resonators, including high-$Q$ cavities and photonic-crystal structures; the common element is that free-space radiation channels are at least partially suppressed by a mirror, cavity wall, or photon-collection chamber (Millar et al., 2016, Alesini et al., 2022).

1. Definition and scope

A dielectric haloscope converts axion dark matter into microwaves by exploiting the axion-induced electric field that appears in the presence of a static magnetic field. In the mirror-terminated formulation, what makes the device “closed” is a reflective boundary, typically a conducting mirror or backplate, that closes one side of the stack, imposes cavity-like boundary conditions on that side, and directs the emission out of the open side. The mirror enforces $E\approx 0$ inside the conductor, so the discontinuity at the mirror–vacuum interface guarantees emission; together with a multi-disk stack, it shapes the mode structure and increases directionality toward the receiver (Group et al., 2016).

The term is not used uniformly. A mirror-backed stack can still be classified as open if the electromagnetic field is emitted into free space and collected by a horn or quasi-optical receiver without a resonant enclosure. A millimeter-wave dielectric haloscope built from four LaAlO$_3$ disks and a mirror was explicitly categorized this way: despite a metal shield and a backing mirror, the photons were emitted into free space toward an antenna, so the instrument was described as an open dielectric haloscope rather than a closed one (Wei et al., 30 Mar 2025).

By contrast, a fully metal-enclosed dielectric cavity is “closed” in the stronger sense that the electromagnetic field is bounded by cavity walls. This includes dielectric-loaded cylindrical resonators, Bragg and DBAS resonators, and photonic-crystal cavities, in which the dielectric elements shape a discrete resonant mode rather than a one-sided free-space emission pattern (Alesini et al., 2022, Bae et al., 2022). The same terminology has also been extended to non-resonant photon-collection chambers, as in DPHaSE, where a dielectric conversion target is enclosed within a photon collection chamber that traps and reprocesses produced photons until they are absorbed on an internal SNSPD (Koppell et al., 30 May 2025).

A common misconception is therefore that “closed dielectric haloscope” denotes a single geometry. The published record instead shows a family of related devices: mirror-terminated boosters, multilayer Fabry–Pérot-like resonators, dielectric-loaded microwave cavities, and enclosed photonic structures.

2. Axion electrodynamics and interface emission

The starting point is the axion–photon interaction,

$$\mathcal{L}\supset -\frac{g_{a\gamma}}{4}F_{\mu\nu}\tilde F^{\mu\nu} a,$$

which modifies Maxwell’s equations in an external static magnetic field. For nonrelativistic halo axions, spatial gradients are negligible on apparatus scales, so $a(t)=a_0\cos(\omega_a t)$ with $\omega_a\simeq m_a$, and the effective source term reduces to

$$\mathbf{J}_a = g_{a\gamma}\,\mathbf{B}_E\,\partial_t a.$$

In a homogeneous medium of permittivity $\epsilon$, the induced electric field is

$$\mathbf{E}_a(t)= -\frac{g_{a\gamma}\mathbf{B}_E}{\epsilon}a(t).$$

Because $\mathbf{E}_a\propto 1/\epsilon$, it is discontinuous across dielectric interfaces; propagating electromagnetic waves must then be generated to satisfy the standard tangential continuity conditions (Millar et al., 2016).

For a planar interface between media 1 and 2 with refractive indices $E\approx 0$0, the emitted wave amplitudes are

$E\approx 0$1

In the perfect-mirror limit, the closed boundary enforces $E\approx 0$2 at the surface, so the outgoing wave into vacuum has amplitude $E\approx 0$3: the axion-induced tangential field must be cancelled locally by radiation, and emission becomes one-sided (Millar et al., 2016).

For multilayer systems, the standard description is a transfer-matrix or transmission-line cascade. In one representation,

$E\approx 0$4

where $E\approx 0$5 propagates through region $E\approx 0$6, $E\approx 0$7 is the interface matrix, and $E\approx 0$8 is the axion source at the interface jump (Millar et al., 2016).

A complementary formulation avoids computing the unknown axion-induced field directly. By Lorentz reciprocity, the signal power can be expressed in terms of measurable reflection-induced fields:

$E\approx 0$9

in the cold-DM limit. This relation applies to resonant cavities, dielectric haloscopes, and broadband dish antennas, and is particularly useful when reflection measurements are experimentally easier than direct field mapping (Egge, 2022).

3. Mirror-terminated multilayer boosters

In the mirror-plus-disk realization, each interface and the mirror act as phased radiators driven by the effective axion current. The boost factor $_3$0 is defined by the emitted field or power relative to the single-mirror baseline, and the one-sided signal power is

$_3$1

For a one-sided closed haloscope, the mirror suppresses backward emission, concentrates the power into one receiver port, and increases directivity without violating the area law (Millar et al., 2016).

Two limiting operating regimes are emphasized. In a resonant cavity-like mode, such as a single disk plus mirror at $_3$2 and $_3$3, the quality factor scales as $_3$4; the peak boost grows with refractive index while the bandwidth shrinks as $_3$5. In a transparent mode with $_3$6, each period adds a $_3$7 propagation phase, so $_3$8 disks add coherently and the peak scales approximately as $_3$9 while the useful bandwidth scales as $$\mathcal{L}\supset -\frac{g_{a\gamma}}{4}F_{\mu\nu}\tilde F^{\mu\nu} a,$$0 (Millar et al., 2016).

The area law governs the trade between bandwidth and peak enhancement. For lossless media, the configuration- or frequency-averaged emitted power depends primarily on the number of interfaces rather than on the detailed spacings, and the integral $$\mathcal{L}\supset -\frac{g_{a\gamma}}{4}F_{\mu\nu}\tilde F^{\mu\nu} a,$$1 is approximately conserved when disk positions are varied. Closing one side preserves this law while redistributing spectral weight and increasing directivity (Millar et al., 2016). Operationally, this means that broadband settings minimize retuning overhead, while narrow resonances maximize peak power and are well suited to rescans (Knirck, 2017).

Representative designs target $$\mathcal{L}\supset -\frac{g_{a\gamma}}{4}F_{\mu\nu}\tilde F^{\mu\nu} a,$$2–$$\mathcal{L}\supset -\frac{g_{a\gamma}}{4}F_{\mu\nu}\tilde F^{\mu\nu} a,$$3, corresponding to $$\mathcal{L}\supset -\frac{g_{a\gamma}}{4}F_{\mu\nu}\tilde F^{\mu\nu} a,$$4–$$\mathcal{L}\supset -\frac{g_{a\gamma}}{4}F_{\mu\nu}\tilde F^{\mu\nu} a,$$5 axion masses. Quoted hardware ranges include magnetic fields of $$\mathcal{L}\supset -\frac{g_{a\gamma}}{4}F_{\mu\nu}\tilde F^{\mu\nu} a,$$6–$$\mathcal{L}\supset -\frac{g_{a\gamma}}{4}F_{\mu\nu}\tilde F^{\mu\nu} a,$$7, aperture $$\mathcal{L}\supset -\frac{g_{a\gamma}}{4}F_{\mu\nu}\tilde F^{\mu\nu} a,$$8–$$\mathcal{L}\supset -\frac{g_{a\gamma}}{4}F_{\mu\nu}\tilde F^{\mu\nu} a,$$9, mm-scale sapphire or LaAlO$a(t)=a_0\cos(\omega_a t)$0 disks, and $a(t)=a_0\cos(\omega_a t)$1–$a(t)=a_0\cos(\omega_a t)$2 disks. Broadband closed stacks with a mirror plus $a(t)=a_0\cos(\omega_a t)$3–$a(t)=a_0\cos(\omega_a t)$4 disks are stated to achieve $a(t)=a_0\cos(\omega_a t)$5–$a(t)=a_0\cos(\omega_a t)$6, with signal powers of order $a(t)=a_0\cos(\omega_a t)$7–$a(t)=a_0\cos(\omega_a t)$8 for $a(t)=a_0\cos(\omega_a t)$9–$\omega_a\simeq m_a$0 and $\omega_a\simeq m_a$1–$\omega_a\simeq m_a$2 across $\omega_a\simeq m_a$3–$\omega_a\simeq m_a$4; for an $\omega_a\simeq m_a$5-disk, $\omega_a\simeq m_a$6, $\omega_a\simeq m_a$7 system, sensitivity to QCD axion models is presented as conceivable (Group et al., 2016).

Prototype work established the underlying control methodology. In a copper-mirror booster with up to five sapphire disks, the measured group delay was sufficiently reproduced by one-dimensional calculations, and the repeatability of the tuning was at the percent level, implying small sensitivity impact for MADMAX-like boost profiles (Egge et al., 2020).

4. Resonant metal-enclosed dielectric cavities

A second major branch of closed dielectric haloscopes uses dielectric elements inside a metallic cavity to reshape and confine a discrete resonant mode. In this setting, the relevant figure of merit is usually written in the cavity form

$\omega_a\simeq m_a$8

with the dielectric structures chosen to increase $\omega_a\simeq m_a$9, $$\mathbf{J}_a = g_{a\gamma}\,\mathbf{B}_E\,\partial_t a.$$0, or both while keeping the mode axion-sensitive (Alesini et al., 2022).

The QUAX–$$\mathbf{J}_a = g_{a\gamma}\,\mathbf{B}_E\,\partial_t a.$$1 experiment is the clearest realized example. Its haloscope is a right-cylindrical OFHC copper cavity loaded with hollow single-crystal sapphire cylinders arranged concentrically along the axis, operated in the TM$$\mathbf{J}_a = g_{a\gamma}\,\mathbf{B}_E\,\partial_t a.$$2 mode inside an $$\mathbf{J}_a = g_{a\gamma}\,\mathbf{B}_E\,\partial_t a.$$3 solenoid at $$\mathbf{J}_a = g_{a\gamma}\,\mathbf{B}_E\,\partial_t a.$$4. The dielectric loading reduces wall participation and magnetoresistive losses, yielding internal $$\mathbf{J}_a = g_{a\gamma}\,\mathbf{B}_E\,\partial_t a.$$5 in field and loaded $$\mathbf{J}_a = g_{a\gamma}\,\mathbf{B}_E\,\partial_t a.$$6 during axion runs. In the interval $$\mathbf{J}_a = g_{a\gamma}\,\mathbf{B}_E\,\partial_t a.$$7–$$\mathbf{J}_a = g_{a\gamma}\,\mathbf{B}_E\,\partial_t a.$$8, corresponding to $$\mathbf{J}_a = g_{a\gamma}\,\mathbf{B}_E\,\partial_t a.$$9, the experiment set a $\epsilon$0 C.L. limit $\epsilon$1 (Alesini et al., 2022).

Super-mode dielectric resonators extend this idea. In the TM$\epsilon$2 DBAS ring resonator, a sapphire ring is placed so that the out-of-phase region of TM$\epsilon$3 is confined in dielectric, suppressing phase cancellation. Simulations give $\epsilon$4 for DBAS TM$\epsilon$5, compared with $\epsilon$6 for an empty-cavity TM$\epsilon$7, and $\epsilon$8 for the TM$\epsilon$9-like DBAS mode. Splitting the ring axially yields symmetric and anti-symmetric super-modes and provides tuning by $$\mathbf{E}_a(t)= -\frac{g_{a\gamma}\mathbf{B}_E}{\epsilon}a(t).$$0 starting from $$\mathbf{E}_a(t)= -\frac{g_{a\gamma}\mathbf{B}_E}{\epsilon}a(t).$$1. Cryogenic Bragg-ring measurements reported $$\mathbf{E}_a(t)= -\frac{g_{a\gamma}\mathbf{B}_E}{\epsilon}a(t).$$2 for a TM$$\mathbf{E}_a(t)= -\frac{g_{a\gamma}\mathbf{B}_E}{\epsilon}a(t).$$3 mode at $$\mathbf{E}_a(t)= -\frac{g_{a\gamma}\mathbf{B}_E}{\epsilon}a(t).$$4 (McAllister et al., 2017).

A related DBAS strategy uses azimuthal wedges to recover axion sensitivity in TM$$\mathbf{E}_a(t)= -\frac{g_{a\gamma}\mathbf{B}_E}{\epsilon}a(t).$$5 modes that would otherwise integrate to zero. In modeled wedge DBAS resonators, form factors up to $$\mathbf{E}_a(t)= -\frac{g_{a\gamma}\mathbf{B}_E}{\epsilon}a(t).$$6 are obtained for TM$$\mathbf{E}_a(t)= -\frac{g_{a\gamma}\mathbf{B}_E}{\epsilon}a(t).$$7 and TM$$\mathbf{E}_a(t)= -\frac{g_{a\gamma}\mathbf{B}_E}{\epsilon}a(t).$$8, and the combined scan-time performance across $$\mathbf{E}_a(t)= -\frac{g_{a\gamma}\mathbf{B}_E}{\epsilon}a(t).$$9–$\mathbf{E}_a\propto 1/\epsilon$0 is stated to outperform a rod-tuned TM$\mathbf{E}_a\propto 1/\epsilon$1 benchmark by a factor $\mathbf{E}_a\propto 1/\epsilon$2 (Quiskamp et al., 2020).

Photonic-crystal haloscopes realize closure differently: the resonance is set primarily by the lattice interspace of a periodic dielectric array rather than by the macroscopic cavity size. A $\mathbf{E}_a\propto 1/\epsilon$3 auxetic-tuned dielectric array inside an OFHC copper cavity was measured at $\mathbf{E}_a\propto 1/\epsilon$4, with continuous tuning from $\mathbf{E}_a\propto 1/\epsilon$5 to $\mathbf{E}_a\propto 1/\epsilon$6 over a $\mathbf{E}_a\propto 1/\epsilon$7 rotation and peak $\mathbf{E}_a\propto 1/\epsilon$8 near $\mathbf{E}_a\propto 1/\epsilon$9. In three-dimensional simulations near $E\approx 0$00, the photonic-crystal design had $E\approx 0$01 and $E\approx 0$02, outperforming wire-metamaterial and multicell reference geometries in $E\approx 0$03 under cryogenic assumptions (Bae et al., 2022).

5. Sensitivity determination, diagnostics, and systematics

Closed dielectric haloscopes are unusually dependent on reflection diagnostics because the same boundaries that enhance the signal also structure the calibration observables. In multilayer Fabry–Pérot-like devices, the group delay

$E\approx 0$04

provides a direct measure of resonance lifetime, with $E\approx 0$05. An echo-free methodology developed for the DALI prototype uses mirror-only normalization and DPSS time-domain gating to isolate the device-under-test response. With YSZ multilayers closed by a copper mirror, this method measured $E\approx 0$06 and $E\approx 0$07 at $E\approx 0$08 for $E\approx 0$09, and found the scaling $E\approx 0$10, with extrapolated $E\approx 0$11 for $E\approx 0$12 at several-dozen-GHz frequencies (Hernández-Cabrera et al., 2024).

Reflection measurements also underpin quantitative signal prediction. The reciprocity formalism relates the axion power to measurable reflection-induced fields, while cavity-style experiments infer $E\approx 0$13, coupling, and mode structure from fitted reflection and transmission spectra. This is especially important when axion excitation and calibration excitation do not populate identical field distributions, a situation that applies to dielectric haloscopes more generally than to conventional cavity haloscopes (Egge, 2022).

Three-dimensional effects impose nontrivial corrections. For ideal finite-diameter dielectric boosters, a geometric form factor can reduce the emitted power by up to $E\approx 0$14 relative to earlier one-dimensional calculations. In the benchmark $E\approx 0$15, $E\approx 0$16 configuration studied for MADMAX, the fundamental mode carries about $E\approx 0$17 of the axion-induced power, the emitted beam is approximately Gaussian with waist $E\approx 0$18, and the design requirements for less than $E\approx 0$19 power change are a maximum disk tilt of $E\approx 0$20 divided by the disk diameter, disk planarity of $E\approx 0$21 (min-to-max) or better, and surface roughness of $E\approx 0$22 (min-to-max) (Knirck et al., 2021).

The 2026 MADMAX sensitivity model shows how these ingredients are combined in a closed cylindrical waveguide implementation. In the CB200 prototype, three $E\approx 0$23 sapphire disks of diameter $E\approx 0$24 are enclosed in an aluminum cylinder of radius $E\approx 0$25 and coupled through a taper to the TE$E\approx 0$26 mode. The fitted transverse overlap is $E\approx 0$27, with an assigned uncertainty $E\approx 0$28, and the measured closed-booster response reaches $E\approx 0$29 up to $E\approx 0$30 near $E\approx 0$31–$E\approx 0$32 in the $E\approx 0$33 Morpurgo magnet. Mode crowding is controlled by choosing the taper-to-first-disk spacing so that higher-order modes are typically $E\approx 0$34–$E\approx 0$35 away from the TE$E\approx 0$36 booster resonance (Ivanov et al., 5 Mar 2026).

The dominant non-idealities recur across architectures: dielectric loss tangent, finite mirror conductivity, disk tilt and non-planarity, finite aperture and diffraction, magnetic-field nonuniformity, and receiver standing waves. The literature does not treat these as secondary corrections; in several implementations, effective fitted parameters such as $E\approx 0$37 and optical thickness are used precisely to absorb residual three-dimensional imperfections into tractable sensitivity models (Millar et al., 2016, Ivanov et al., 5 Mar 2026).

6. Experimental status and directions of development

The closed dielectric haloscope program now spans theory, component validation, and first searches. The original dielectric-haloscope proposal targeted the $E\approx 0$38–$E\approx 0$39 range with up to $E\approx 0$40 disks of area $E\approx 0$41 in a $E\approx 0$42 field, and presented a three-year quantum-limited scenario with discovery potential down to $E\approx 0$43–$E\approx 0$44 over roughly $E\approx 0$45–$E\approx 0$46, extendable to $E\approx 0$47 by adding disks or run time (Group et al., 2016).

On the resonant side, QUAX has already demonstrated that a metal-enclosed dielectric cavity can operate with $E\approx 0$48 in an $E\approx 0$49 magnet at $E\approx 0$50 and set a concrete axion-photon limit near $E\approx 0$51 (Alesini et al., 2022). Photonic-crystal and DBAS resonators, while not yet used for comparable exclusion results, have shown experimentally that high-$E\approx 0$52, tunable, dielectric-shaped modes can be sustained around $E\approx 0$53 and above, with scan-rate and form-factor advantages over conventional rod-tuned geometries in the high-mass regime (Bae et al., 2022, McAllister et al., 2017).

For multilayer mirror-terminated systems, the most advanced closed sensitivity framework is now the MADMAX CB200 model, which explicitly includes realistic geometric imperfections and receiver noise and is described as the foundation for the first axion dark matter search using a dielectric haloscope (Ivanov et al., 5 Mar 2026). In parallel, the echo-free DALI program projects $E\approx 0$54 for $E\approx 0$55 and a reach of $E\approx 0$56 over $E\approx 0$57–$E\approx 0$58 for a full-scale instrument (Hernández-Cabrera et al., 2024).

The main architectural trade-off remains the bandwidth–enhancement balance. Closing the structure simplifies collection, reduces the number of receiver ports, and strengthens directivity or resonant buildup, but it also narrows usable bandwidth and makes mode management, alignment, and calibration central. This suggests that “closed dielectric haloscope” is best understood not as a single detector topology but as a design space in which mirrors, conducting enclosures, dielectric patterning, and calibrated receiver coupling are combined to preserve large conversion volume at frequencies where conventional cavity scaling becomes restrictive.