Hyperbolic Phonon Polaritons in Natural Crystals

• Hyperbolic phonon polaritons are hybrid electromagnetic–lattice excitations in anisotropic crystals, characterized by opposite-sign dielectric tensor components that allow high in-plane momenta and deep subwavelength confinement.
• They naturally occur in materials such as hexagonal boron nitride and α-MoO₃, enabling applications like nanocavity formation, beam steering, and enhanced local sensing.
• Substrate and geometry engineering dynamically tune their dispersion, confinement, and energy transport, paving the way for advanced mid-IR nanophotonic devices.

Searching arXiv for recent and foundational papers on hyperbolic phonon polaritons to ground the article in the literature. Hyperbolic phonon polaritons are hybrid electromagnetic–lattice excitations sustained by polar anisotropic crystals in spectral intervals where principal components of the dielectric tensor have opposite signs. In these Reststrahlen bands, the isofrequency surfaces become open hyperboloids rather than closed ellipsoids, enabling deeply subwavelength momenta, directional energy flow, and guided or volume-confined infrared modes with strong field localization. In natural van der Waals crystals such as hexagonal boron nitride and biaxial oxides such as $\alpha$-MoO$_3$, these modes appear without artificial metamaterial structuring and have been exploited for nanocavity formation, beam steering, substrate-controlled dispersion engineering, long-range energy transfer, nanofocusing, and subsurface diagnosis (Barcelos et al., 2021, Heiden et al., 10 Apr 2026, Álvarez-Pérez et al., 6 Mar 2026, Hutchins et al., 2024, Dai et al., 2018).

1. Definition, dielectric criteria, and hyperbolic dispersion

Phonon polaritons arise from coupling between infrared photons and optical phonons in polar crystals. Within the interval between transverse optical and longitudinal optical phonon frequencies, the real part of the permittivity becomes negative, allowing strongly confined electromagnetic modes. In anisotropic crystals, the permittivity is tensorial, for example

$$\boldsymbol{\varepsilon}(\omega)= \begin{pmatrix} \varepsilon_{xx}(\omega) & 0 & 0 \ 0 & \varepsilon_{yy}(\omega) & 0 \ 0 & 0 & \varepsilon_{zz}(\omega) \end{pmatrix},$$

and hyperbolicity occurs when the real parts of at least two components have opposite signs, such as

$$\operatorname{Re}[\varepsilon_i(\omega)]\cdot \operatorname{Re}[\varepsilon_j(\omega)] < 0.$$

This sign-indefinite response converts the momentum-space isofrequency surface into a hyperboloid and permits very large in-plane wavevectors relative to the free-space value (Barcelos et al., 2021, Heiden et al., 10 Apr 2026).

For uniaxial media such as hBN, extraordinary-wave dispersion is commonly written as

$$\frac{k_x^2 + k_y^2}{\varepsilon_{\parallel}(\omega)} + \frac{k_z^2}{\varepsilon_{\perp}(\omega)} = \left(\frac{\omega}{c}\right)^2$$

or, equivalently under the notation used for hBN,

$$\frac{k_x^2 + k_y^2}{\varepsilon_z(\omega)} + \frac{k_z^2}{\varepsilon_t(\omega)} = \left(\frac{\omega}{c}\right)^2.$$

For biaxial systems such as $\alpha$-MoO$_3$, in-plane propagation can be described approximately by

$$\frac{k_x^2}{\varepsilon_y(\omega)} + \frac{k_y^2}{\varepsilon_x(\omega)} = \left(\frac{\omega}{c}\right)^2,$$

so the orientation of the open contour depends directly on the signs and magnitudes of $\varepsilon_x$ and $_3$0 (Barcelos et al., 2021, Zheng et al., 2021).

Several consequences follow from this dispersion. Hyperbolic phonon polaritons support large in-plane momenta $_3$1, strong confinement, directional or ray-like propagation, and ultra-slow group velocities with long lifetimes because the underlying material excitation is phononic rather than electronic (Barcelos et al., 2021). In hBN, HPhP wavelengths can reach $_3$2–$_3$3 nm while free-space wavelengths are of order $_3$4, evidencing wavelength compression by factors of $_3$5–$_3$6 in the relevant bands (Tomadin et al., 2015). In $_3$7-MoO$_3$8, confinement factors around $_3$9 were reported at $$\boldsymbol{\varepsilon}(\omega)= \begin{pmatrix} \varepsilon_{xx}(\omega) & 0 & 0 \ 0 & \varepsilon_{yy}(\omega) & 0 \ 0 & 0 & \varepsilon_{zz}(\omega) \end{pmatrix},$$0 in nanobelt cavities (Barcelos et al., 2021), while plasmonic-antenna control produced $$\boldsymbol{\varepsilon}(\omega)= \begin{pmatrix} \varepsilon_{xx}(\omega) & 0 & 0 \ 0 & \varepsilon_{yy}(\omega) & 0 \ 0 & 0 & \varepsilon_{zz}(\omega) \end{pmatrix},$$1 at $$\boldsymbol{\varepsilon}(\omega)= \begin{pmatrix} \varepsilon_{xx}(\omega) & 0 & 0 \ 0 & \varepsilon_{yy}(\omega) & 0 \ 0 & 0 & \varepsilon_{zz}(\omega) \end{pmatrix},$$2 (Zheng et al., 2021).

A common misconception is that hyperbolicity is synonymous with merely having a negative permittivity. The literature distinguishes ordinary phonon polaritons from hyperbolic phonon polaritons by the sign opposition among tensor components, not by negativity alone. Another common simplification is to treat the dielectric tensor as dependent only on electric-field polarization. A more recent analysis argued that once LO–TO splitting is fully included, dielectric functions become wave-vector-direction dependent as well as polarization dependent, improving agreement with measured hBN permittivity and predicting unusual dumbbell-shaped and butterfly-shaped isofrequency curves in some hexagonal crystals (Fang et al., 2024).

2. Natural material platforms and Reststrahlen-band taxonomy

Natural HPhP platforms span uniaxial and biaxial polar crystals. Hexagonal boron nitride is the canonical uniaxial system, with dielectric tensor

$$\boldsymbol{\varepsilon}(\omega)= \begin{pmatrix} \varepsilon_{xx}(\omega) & 0 & 0 \ 0 & \varepsilon_{yy}(\omega) & 0 \ 0 & 0 & \varepsilon_{zz}(\omega) \end{pmatrix},$$3

Its lower and upper Reststrahlen bands correspond to opposite sign configurations of $$\boldsymbol{\varepsilon}(\omega)= \begin{pmatrix} \varepsilon_{xx}(\omega) & 0 & 0 \ 0 & \varepsilon_{yy}(\omega) & 0 \ 0 & 0 & \varepsilon_{zz}(\omega) \end{pmatrix},$$4 and $$\boldsymbol{\varepsilon}(\omega)= \begin{pmatrix} \varepsilon_{xx}(\omega) & 0 & 0 \ 0 & \varepsilon_{yy}(\omega) & 0 \ 0 & 0 & \varepsilon_{zz}(\omega) \end{pmatrix},$$5, yielding two hyperbolic windows in the mid-infrared (Heiden et al., 10 Apr 2026, Ambrosio et al., 2017). In the terminology of one hBN study, the upper Reststrahlen band $$\boldsymbol{\varepsilon}(\omega)= \begin{pmatrix} \varepsilon_{xx}(\omega) & 0 & 0 \ 0 & \varepsilon_{yy}(\omega) & 0 \ 0 & 0 & \varepsilon_{zz}(\omega) \end{pmatrix},$$6 is a type II regime with $$\boldsymbol{\varepsilon}(\omega)= \begin{pmatrix} \varepsilon_{xx}(\omega) & 0 & 0 \ 0 & \varepsilon_{yy}(\omega) & 0 \ 0 & 0 & \varepsilon_{zz}(\omega) \end{pmatrix},$$7 and $$\boldsymbol{\varepsilon}(\omega)= \begin{pmatrix} \varepsilon_{xx}(\omega) & 0 & 0 \ 0 & \varepsilon_{yy}(\omega) & 0 \ 0 & 0 & \varepsilon_{zz}(\omega) \end{pmatrix},$$8 (Ambrosio et al., 2017). Another study reported type I and type II hyperbolic bands for hBN as $$\boldsymbol{\varepsilon}(\omega)= \begin{pmatrix} \varepsilon_{xx}(\omega) & 0 & 0 \ 0 & \varepsilon_{yy}(\omega) & 0 \ 0 & 0 & \varepsilon_{zz}(\omega) \end{pmatrix},$$9 and $$\operatorname{Re}[\varepsilon_i(\omega)]\cdot \operatorname{Re}[\varepsilon_j(\omega)] < 0.$$0, respectively (Fang et al., 2024).

$$\operatorname{Re}[\varepsilon_i(\omega)]\cdot \operatorname{Re}[\varepsilon_j(\omega)] < 0.$$1-MoO$$\operatorname{Re}[\varepsilon_i(\omega)]\cdot \operatorname{Re}[\varepsilon_j(\omega)] < 0.$$2 is a biaxial van der Waals semiconductor with crystal axes $$\operatorname{Re}[\varepsilon_i(\omega)]\cdot \operatorname{Re}[\varepsilon_j(\omega)] < 0.$$3, $$\operatorname{Re}[\varepsilon_i(\omega)]\cdot \operatorname{Re}[\varepsilon_j(\omega)] < 0.$$4, and $$\operatorname{Re}[\varepsilon_i(\omega)]\cdot \operatorname{Re}[\varepsilon_j(\omega)] < 0.$$5. Because $$\operatorname{Re}[\varepsilon_i(\omega)]\cdot \operatorname{Re}[\varepsilon_j(\omega)] < 0.$$6, $$\operatorname{Re}[\varepsilon_i(\omega)]\cdot \operatorname{Re}[\varepsilon_j(\omega)] < 0.$$7, and $$\operatorname{Re}[\varepsilon_i(\omega)]\cdot \operatorname{Re}[\varepsilon_j(\omega)] < 0.$$8 change sign at different phonon resonances, $$\operatorname{Re}[\varepsilon_i(\omega)]\cdot \operatorname{Re}[\varepsilon_j(\omega)] < 0.$$9-MoO$$\frac{k_x^2 + k_y^2}{\varepsilon_{\parallel}(\omega)} + \frac{k_z^2}{\varepsilon_{\perp}(\omega)} = \left(\frac{\omega}{c}\right)^2$$0 hosts multiple Reststrahlen bands across the far-IR and mid-IR. One report describes four distinct bands, RB1–RB4, featuring type I and type II behavior along different axes and frequency-dependent transitions between in-plane and out-of-plane hyperbolicity (Barcelos et al., 2021). Another gives three Reststrahlen bands in the $$\frac{k_x^2 + k_y^2}{\varepsilon_{\parallel}(\omega)} + \frac{k_z^2}{\varepsilon_{\perp}(\omega)} = \left(\frac{\omega}{c}\right)^2$$1 range, with Band 2 characterized by $$\frac{k_x^2 + k_y^2}{\varepsilon_{\parallel}(\omega)} + \frac{k_z^2}{\varepsilon_{\perp}(\omega)} = \left(\frac{\omega}{c}\right)^2$$2, $$\frac{k_x^2 + k_y^2}{\varepsilon_{\parallel}(\omega)} + \frac{k_z^2}{\varepsilon_{\perp}(\omega)} = \left(\frac{\omega}{c}\right)^2$$3, and Band 3 by $$\frac{k_x^2 + k_y^2}{\varepsilon_{\parallel}(\omega)} + \frac{k_z^2}{\varepsilon_{\perp}(\omega)} = \left(\frac{\omega}{c}\right)^2$$4, $$\frac{k_x^2 + k_y^2}{\varepsilon_{\parallel}(\omega)} + \frac{k_z^2}{\varepsilon_{\perp}(\omega)} = \left(\frac{\omega}{c}\right)^2$$5 (Zheng et al., 2021). These descriptions are consistent in emphasizing that $$\frac{k_x^2 + k_y^2}{\varepsilon_{\parallel}(\omega)} + \frac{k_z^2}{\varepsilon_{\perp}(\omega)} = \left(\frac{\omega}{c}\right)^2$$6-MoO$$\frac{k_x^2 + k_y^2}{\varepsilon_{\parallel}(\omega)} + \frac{k_z^2}{\varepsilon_{\perp}(\omega)} = \left(\frac{\omega}{c}\right)^2$$7 supports both in-plane and out-of-plane hyperbolic regimes, with especially strong in-plane anisotropy (Zheng et al., 2021).

Several materials comparisons recur in the literature. hBN is a natural low-loss hyperbolic crystal with two principal hyperbolic bands (Barcelos et al., 2021). $$\frac{k_x^2 + k_y^2}{\varepsilon_{\parallel}(\omega)} + \frac{k_z^2}{\varepsilon_{\perp}(\omega)} = \left(\frac{\omega}{c}\right)^2$$8-MoO$$\frac{k_x^2 + k_y^2}{\varepsilon_{\parallel}(\omega)} + \frac{k_z^2}{\varepsilon_{\perp}(\omega)} = \left(\frac{\omega}{c}\right)^2$$9 offers more spectral windows, in-plane hyperbolicity, and biaxial control (Barcelos et al., 2021, Zheng et al., 2021). SiC is cited as a bulk polar crystal with a single Reststrahlen band and relatively isotropic behavior unless structured (Barcelos et al., 2021). A broader survey of hexagonal compounds concluded that, besides hBN, h-AlN exhibits a wide hyperbolic frequency-band range, whereas h-BP, h-AlP, h-GaN, and h-GaP show it scarcely (Fang et al., 2024).

This suggests that the width and utility of a hyperbolic band depend not only on anisotropy, but on the separation of infrared-active TO frequencies along different directions, phonon lifetime, and overall phonon energy scale (Fang et al., 2024). That conclusion is drawn explicitly in the h-AlN comparison study rather than inferred from general intuition.

3. Mode structure, confinement, and dispersion orders

In finite slabs, HPhPs are quantized into thickness-dependent branches or dispersion orders. For hBN slabs, one study writes the in-plane momentum of branch $$\frac{k_x^2 + k_y^2}{\varepsilon_z(\omega)} + \frac{k_z^2}{\varepsilon_t(\omega)} = \left(\frac{\omega}{c}\right)^2.$$0 as

$$\frac{k_x^2 + k_y^2}{\varepsilon_z(\omega)} + \frac{k_z^2}{\varepsilon_t(\omega)} = \left(\frac{\omega}{c}\right)^2.$$1

where $$\frac{k_x^2 + k_y^2}{\varepsilon_z(\omega)} + \frac{k_z^2}{\varepsilon_t(\omega)} = \left(\frac{\omega}{c}\right)^2.$$2 is slab thickness, $$\frac{k_x^2 + k_y^2}{\varepsilon_z(\omega)} + \frac{k_z^2}{\varepsilon_t(\omega)} = \left(\frac{\omega}{c}\right)^2.$$3 is the mode order, $$\frac{k_x^2 + k_y^2}{\varepsilon_z(\omega)} + \frac{k_z^2}{\varepsilon_t(\omega)} = \left(\frac{\omega}{c}\right)^2.$$4 is the substrate permittivity, and $$\frac{k_x^2 + k_y^2}{\varepsilon_z(\omega)} + \frac{k_z^2}{\varepsilon_t(\omega)} = \left(\frac{\omega}{c}\right)^2.$$5 encode wavelength and damping (Noh et al., 23 Jan 2026). In this framework, the $$\frac{k_x^2 + k_y^2}{\varepsilon_z(\omega)} + \frac{k_z^2}{\varepsilon_t(\omega)} = \left(\frac{\omega}{c}\right)^2.$$6 mode is the fundamental branch, while $$\frac{k_x^2 + k_y^2}{\varepsilon_z(\omega)} + \frac{k_z^2}{\varepsilon_t(\omega)} = \left(\frac{\omega}{c}\right)^2.$$7 are higher-order branches with incrementally higher momenta and different parity. Finite-difference frequency-domain calculations in the same work show even and odd $$\frac{k_x^2 + k_y^2}{\varepsilon_z(\omega)} + \frac{k_z^2}{\varepsilon_t(\omega)} = \left(\frac{\omega}{c}\right)^2.$$8 field profiles across thickness for different $$\frac{k_x^2 + k_y^2}{\varepsilon_z(\omega)} + \frac{k_z^2}{\varepsilon_t(\omega)} = \left(\frac{\omega}{c}\right)^2.$$9, establishing a modal taxonomy based on vertical nodal structure (Noh et al., 23 Jan 2026).

For thin hBN slabs over metal, gap engineering can transform a fundamental free-standing HPhP continuously into an image HPhP. At $\alpha$0, a $\alpha$1 nm hBN slab above Au evolves from an “almost suspended” mode at $\alpha$2 nm to a strongly confined image HPhP at $\alpha$3 nm, with the in-plane momentum increasing by more than three times relative to the free-standing case at mid-band (Heiden et al., 10 Apr 2026). Experimentally, for the same $\alpha$4 nm slab, $\alpha$5 increases by about $\alpha$6 as the gap changes from $\alpha$7 nm to $\alpha$8 nm over $\alpha$9 (Heiden et al., 10 Apr 2026).

For $_3$0-MoO$_3$1, slab-guided HPhPs in lithography-free nanobelts are quantized by the belt width, producing Fabry–Perot modes across several Reststrahlen bands (Barcelos et al., 2021). The cavity condition is written approximately as

$_3$2

where $_3$3 is the effective cavity length, $_3$4 is the edge reflection phase, and $_3$5 is the resonance order (Barcelos et al., 2021). The experimentally imaged standing-wave maxima correspond to different integer orders $_3$6, and the extracted in-plane momenta lie on the dispersion of the fundamental guided mode $_3$7 of an infinite slab of the same thickness (Barcelos et al., 2021).

An important correction to a common oversimplification is that different HPhP orders are not merely higher harmonics of one in-plane wave. They correspond to distinct slab eigenmodes with different field symmetry and thickness quantization (Noh et al., 23 Jan 2026). Another misconception is that higher order always implies stronger environmental sensitivity. A substrate-index study found the opposite in relative terms: the absolute change $_3$8 under substrate modification is nearly the same across HPhP orders, but the fractional change is smaller for higher orders because they begin at larger baseline $_3$9 (Fali et al., 2019).

4. Launching, imaging, and spectroscopy

Because HPhPs possess momenta far larger than those of free-space photons, excitation and detection require momentum-matching strategies. The standard approach is scattering-type scanning near-field optical microscopy, in which a metallic AFM tip launches and detects polaritons locally. This technique has been used extensively in hBN, $$\frac{k_x^2}{\varepsilon_y(\omega)} + \frac{k_y^2}{\varepsilon_x(\omega)} = \left(\frac{\omega}{c}\right)^2,$$0-MoO$$\frac{k_x^2}{\varepsilon_y(\omega)} + \frac{k_y^2}{\varepsilon_x(\omega)} = \left(\frac{\omega}{c}\right)^2,$$1, and related systems (Ambrosio et al., 2017, Barcelos et al., 2021, Zheng et al., 2021, Heiden et al., 10 Apr 2026, Noh et al., 23 Jan 2026).

In s-SNOM, the tip acts as a nanoscale antenna, supplying high-$$\frac{k_x^2}{\varepsilon_y(\omega)} + \frac{k_y^2}{\varepsilon_x(\omega)} = \left(\frac{\omega}{c}\right)^2,$$2 Fourier components to couple incident mid-IR radiation into HPhPs. Reflections from edges, steps, terraces, or internal defects generate interference fringes whose spacing gives the polariton wavelength. For typical tip-launched standing-wave configurations, the fringe spacing is $$\frac{k_x^2}{\varepsilon_y(\omega)} + \frac{k_y^2}{\varepsilon_x(\omega)} = \left(\frac{\omega}{c}\right)^2,$$3; for edge-launched propagation interfered with a background field, it can equal $$\frac{k_x^2}{\varepsilon_y(\omega)} + \frac{k_y^2}{\varepsilon_x(\omega)} = \left(\frac{\omega}{c}\right)^2,$$4 (Heiden et al., 10 Apr 2026). In hBN, a representative tip-launched wavelength at $$\frac{k_x^2}{\varepsilon_y(\omega)} + \frac{k_y^2}{\varepsilon_x(\omega)} = \left(\frac{\omega}{c}\right)^2,$$5 was reported as $$\frac{k_x^2}{\varepsilon_y(\omega)} + \frac{k_y^2}{\varepsilon_x(\omega)} = \left(\frac{\omega}{c}\right)^2,$$6 nm, matching a calculated value of $$\frac{k_x^2}{\varepsilon_y(\omega)} + \frac{k_y^2}{\varepsilon_x(\omega)} = \left(\frac{\omega}{c}\right)^2,$$7 nm (Woessner et al., 2017).

Several complementary spectroscopic and imaging schemes have expanded the HPhP toolbox. Broadband synchrotron infrared nanospectroscopy combined with pseudo-heterodyne s-SNOM was used to map HPhPs in $$\frac{k_x^2}{\varepsilon_y(\omega)} + \frac{k_y^2}{\varepsilon_x(\omega)} = \left(\frac{\omega}{c}\right)^2,$$8-MoO$$\frac{k_x^2}{\varepsilon_y(\omega)} + \frac{k_y^2}{\varepsilon_x(\omega)} = \left(\frac{\omega}{c}\right)^2,$$9 nanobelts from $\varepsilon_x$0 to $\varepsilon_x$1, while tunable QCL-based s-SNOM provided higher-SNR narrowband images for mode assignment (Barcelos et al., 2021). Mechanical detection via AFM contact-mode cantilever oscillations was shown to image hBN HPhPs without detecting scattered photons, using photothermal expansion induced by locally enhanced lattice vibrations (Ambrosio et al., 2017). That method achieved better than $\varepsilon_x$2 resolution and reproduced the same fringe periodicity as s-NSOM on the same flakes (Ambrosio et al., 2017).

Far-field access has historically been difficult because of the large momentum mismatch. Two distinct solutions emerge in the literature. One is electrical detection in graphene/hBN heterostructures, where split metallic gates launch HPPs in hBN that propagate as confined rays toward a graphene $\varepsilon_x$3-junction, producing photocurrent via hot-carrier photothermoelectric conversion (Woessner et al., 2017). The other is the construction of photonic hypercrystals in $\varepsilon_x$4-MoO$\varepsilon_x$5: periodic nanohole lattices provide reciprocal lattice vectors that phase-match free-space photons to in-plane HPhPs, enabling direct far-field reflectance signatures of hyperbolic modes and twist-controlled resonance tuning (Sahoo et al., 2023).

A further diagnostic route is purely electronic. In graphene/hBN heterostructures, coupling between graphene plasmons and hBN slab phonon polaritons produces plasmon–phonon polariton branches whose signatures appear in angle-resolved photoemission spectra as satellite features and self-energy peaks (Tomadin et al., 2015). This does not detect HPhPs optically; rather, it transfers their spectral fingerprints into graphene’s electronic spectral function.

5. Substrate, cavity, and geometry control

Environmental engineering is one of the dominant mechanisms for controlling HPhPs. In hBN on corrugated gold, a sinusoidally varying gap $\varepsilon_x$6 from $\varepsilon_x$7 to $\varepsilon_x$8 nm produces continuous local tuning of the HPhP wavelength. For $\varepsilon_x$9 nm hBN in the upper Reststrahlen band, $_3$00 increases by about $_3$01 between $_3$02 nm and $_3$03 nm (Heiden et al., 10 Apr 2026). The same structure functions as a “polaritonic taper,” compressing $_3$04 from about $_3$05 nm to about $_3$06 nm at $_3$07, corresponding to about $_3$08 lateral nanofocusing (Heiden et al., 10 Apr 2026).

Substrate refractive index also provides strong control even without metallic image-mode formation. For hBN on suspended, dielectric, and metallic substrates, the thickness-normalized wavevector $_3$09 was shown to vary strongly with substrate permittivity. The wavevector can be reduced by a factor of $_3$10 simply by changing the substrate from dielectric to metallic behavior (Fali et al., 2019). Incorporating the imaginary part of the substrate dielectric function showed that small local variations in loss or carrier density can dynamically control the wavevector and may spatially separate hyperbolic modes of different orders (Fali et al., 2019).

In $_3$11-MoO$_3$12, gradient suspension supplies a different environmental degree of freedom. A wedge-shaped air gap under a suspended flake modifies the Fabry–Perot phase condition for guided HPhPs. The dependence of $_3$13 on gap thickness is opposite in the lower and upper bands: in the in-plane hyperbolic band with $_3$14, $_3$15 increases with gap; in the upper band with $_3$16, $_3$17 decreases with gap (Zheng et al., 2021). The reported tuning reached polariton wavelength elongation up to $_3$18 and damping-rate reduction up to $_3$19 (Zheng et al., 2021).

Geometry on the scale of the polariton wavelength is equally important. Naturally grown quasi-1D $_3$20-MoO$_3$21 nanobelts form lithography-free Fabry–Perot cavities that support HPhP standing waves in multiple bands (Barcelos et al., 2021). Geometrically designed Au antennas on $_3$22-MoO$_3$23 impose a spatial phase profile on launched in-plane HPhPs, enabling frequency-dependent focusing. At $_3$24, a convex antenna extremity yields a focal spot of about $_3$25, around $_3$26 smaller than the free-space wavelength, and at $_3$27 the lateral confinement reaches about $_3$28 (Zheng et al., 2021). The focusing behavior reverses between hyperbolic and elliptic bands, which clarifies that this is not a generic antenna effect but a manifestation of in-plane hyperbolicity (Zheng et al., 2021).

Mode conversion provides a related geometric control principle. In step-shaped hBN and $_3$29-MoO$_3$30 terraces, simple slab edges preserve HPhP order, while asymmetric vdW steps break symmetry and supply additional scattering momentum, enabling $_3$31 conversion and short-period beating patterns observed by s-SNOM (Noh et al., 23 Jan 2026). In hBN terraces at $_3$32, the converted first-order branch produced Fourier peaks near $_3$33, far above the $_3$34 peaks near $_3$35 and $_3$36 (Noh et al., 23 Jan 2026).

6. Directionality, refraction, focusing, and transport

Directional propagation is one of the defining operational features of HPhPs. In hBN, the propagation angle for type-II hyperbolic modes is expressed in one study as

$_3$37

yielding fixed ray trajectories within the slab (Ambrosio et al., 2017). In $_3$38-MoO$_3$39, the in-plane propagation direction depends strongly on frequency and band because different tensor components become negative in different spectral intervals (Barcelos et al., 2021, Zheng et al., 2021). Simulations and near-field maps show propagation elongated along $_3$40 in one band and along $_3$41 in another, directly reflecting the open orientation of the in-plane hyperbola (Barcelos et al., 2021).

This directional control underlies several wave phenomena. Negative refraction was visualized at interfaces between the natural hyperbolic crystals h$_3$42BN and MoO$_3$43, where type-I and type-II phonon polaritons coexist in the same $_3$44 spectral window (Sternbach et al., 2022). At a special frequency $_3$45, the resulting collimated rays circulate along closed diamond-shaped trajectories, and the eigenmode structure shows regions of both positive and negative dispersion separated by gaps associated with polaritonic level repulsion and strong coupling (Sternbach et al., 2022). This indicates that interface engineering between different natural hyperbolic crystals can realize ray optics not available in a single medium.

Directional propagation also drives long-range interaction phenomena. A theoretical framework for $_3$46-MoO$_3$47 slabs showed that hyperbolic phonon polaritons can mediate dipole–dipole interactions with enhancement factors of order $_3$48–$_3$49 along hyperbolic asymptotes, extending significant coupling to $_3$50, about $_3$51, in the mid-IR (Álvarez-Pérez et al., 6 Mar 2026). At $_3$52, enhancement peaks around $_3$53 were reported along the hyperbolic asymptotes; at $_3$54, enhancement remained strong at about $_3$55; and at $_3$56, loss-induced canalization reduced the peak to about $_3$57 while maximizing propagation length and directional transport (Álvarez-Pérez et al., 6 Mar 2026). Twisted bilayers allowed a trade-off between enhancement and directionality, with $_3$58 yielding canalization and $_3$59 giving weaker, more isotropic coupling (Álvarez-Pérez et al., 6 Mar 2026).

Thermal transport provides another example of directional and high-speed energy flow. In hBN, pump–probe thermoreflectance measurements showed that volume-confined HPhP modes excited by near-field radiation from hot Au can mediate interfacial heat transfer with an effective thermal boundary conductance of at least $_3$60, compared to a phonon-only Au/hBN conductance of $_3$61 measured by TDTR (Hutchins et al., 2024). The same study cites HPhP group velocities approaching $_3$62 and a theoretical upper-limit conductance around $_3$63 (Hutchins et al., 2024). A plausible implication is that HPhPs can function as an ultrafast photonic heat channel inside solids, not merely as near-field spectral resonances.

7. Diagnostics, applications, and emerging quantum directions

HPhPs have become a metrological tool as much as a subject of fundamental optics. In hBN, buried defects act as internal HPhP reflectors. By analyzing fringe spacing, reflection coefficient, and dispersion near an internal air gap, one study reconstructed a concealed defect at depth $_3$64 nm below the top surface with thickness $_3$65 nm in a $_3$66 nm slab (Dai et al., 2018). This was achieved from measured polariton wavelengths such as $_3$67 nm at $_3$68 and $_3$69 nm at $_3$70, combined with electromagnetic simulations (Dai et al., 2018). The same work analyzed reflection, transmission, and scattering versus defect size, showing stronger reflection for thicker defects and at higher frequencies because the defect becomes more comparable to the polariton wavelength (Dai et al., 2018).

Photodetection offers a direct device-level application. In graphene/hBN heterostructures above split metallic gates, the gate gap launches HPPs that propagate as confined rays and heat graphene at a $_3$71-junction, producing a photo-thermoelectric photocurrent (Woessner et al., 2017). The external responsivity reached about $_3$72 at room temperature and zero bias, while the internal responsivity was inferred to be about $_3$73 with a noise-equivalent power of about $_3$74 (Woessner et al., 2017). The spectral peak can be tuned over about $_3$75 by varying hBN thickness and split-gate gap width (Woessner et al., 2017).

Far-field optics is becoming increasingly accessible. Patterned $_3$76-MoO$_3$77 photonic hypercrystals convert free-space photons into in-plane HPhPs via reciprocal lattice vectors, producing polarization-selective reflectance resonances in both Reststrahlen bands (Sahoo et al., 2023). Rotating the square hole lattice relative to the $_3$78-MoO$_3$79 crystal axes tunes the far-field resonances because the rotated reciprocal vectors sample different directions of the anisotropic HPhP dispersion (Sahoo et al., 2023). This provides a direct far-field route to probing in-plane hyperbolicity without s-SNOM.

Quantum-optical directions have also emerged. A cavity-QED treatment of hBN color centers proposed two HPP generation schemes: spontaneous emission into the phonon sideband and a stimulated Raman process (Feng et al., 5 Feb 2026). In ultrathin slabs, spontaneous emission becomes effectively single-mode with enhanced decay into a selected HPP branch, while the Raman process provides tunable and narrowband excitation of ray-like HPPs that propagate over micrometer distances (Feng et al., 5 Feb 2026). The work further outlines a two-emitter correlation measurement to test the single-polariton character of these emissions (Feng et al., 5 Feb 2026). This suggests that HPPs may become not only classical nanophotonic carriers but also quantum channels coupling spatially separated emitters.

A recurring misconception is that HPhPs are purely a near-field curiosity. The literature now includes electrically detected HPP-guided photodetectors (Woessner et al., 2017), far-field hypercrystal coupling (Sahoo et al., 2023), ultrafast thermal transport (Hutchins et al., 2024), and quantum-emitter-based source proposals (Feng et al., 5 Feb 2026). Another misconception is that all HPhP functionality comes from nanofabricated resonators. Naturally grown $_3$80-MoO$_3$81 nanobelts already behave as ultrabroadband HPhP nanocavities, demonstrating that crystallographic growth alone can define usable cavity geometries (Barcelos et al., 2021).

Taken together, these results place hyperbolic phonon polaritons at the center of a broad mid-IR and far-IR photonics program: deeply subwavelength waveguiding, cavity formation, local and nonlocal energy transfer, directional beam engineering, environmental sensing, defect tomography, and potentially quantum transduction. The unifying principle is the same in every case: indefinite anisotropic permittivity creates a hyperbolic phase space of high-$_3$82 states, and that phase space can be controlled through thickness, substrate, gap, twist, edge geometry, or heterointerface design (Barcelos et al., 2021, Heiden et al., 10 Apr 2026, Álvarez-Pérez et al., 6 Mar 2026, Noh et al., 23 Jan 2026).