Exceptional Flat Bands Overview

• Exceptional flat bands are dispersionless manifolds characterized by a vanishing group velocity and a high degeneracy of states in both Hermitian and non-Hermitian settings.
• In non-Hermitian systems, they often emerge at exceptional points where eigenvalues and eigenvectors coalesce, enabling tunable lifetimes and unique localization phenomena.
• Robust formation mechanisms—from interference-based compact localized states to symmetry enforcement and diffraction effects—pave the way for applications in photonics, quantum materials, and electronic systems.

Exceptional flat bands are dispersionless spectral manifolds whose defining features depend on context. In the strict non-Hermitian usage, they are flat bands tied to exceptional points (EPs), where eigenvalues and eigenvectors coalesce and the Bloch Hamiltonian becomes defective; in broader usage, the same phrase can denote unusually robust, angle-independent, or symmetry-enforced flat dispersions in Hermitian crystals, wave networks, and radiatively coupled photonic lattices. Across these settings, the common invariants are vanishing group velocity, a large degeneracy of single-particle states, and localization mechanisms ranging from destructive-interference compact localized states (CLS) to moiré-induced velocity collapse and diffraction-selected radiative modes (Biesenthal et al., 2019, Esparza et al., 14 Aug 2025, Lehikoinen et al., 25 Feb 2026).

1. Terminological scope and defining criteria

A flat band is a band with energy independent of quasimomentum, $E(k)=E_{\mathrm{flat}}$, so that the group velocity vanishes identically. In lattice systems this usually implies an extensive degenerate manifold and enables exact or approximate localized eigenmodes. In the canonical interference-based picture, amplitudes and phases are arranged so that hopping out of a finite motif cancels, producing a CLS. In non-Hermitian settings, the same flatness can coexist with complex energies, biorthogonal eigenspaces, and defectiveness at EPs (Biesenthal et al., 2019, Leykam et al., 2017).

The phrase “exceptional flat band” is not used uniformly. In non-Hermitian photonics and non-Hermitian bipartite-crystal theory, it denotes a flat band appearing at, or persisting beyond, an EP. In other literatures, “exceptional” instead refers to exactness, robustness, or a distinctive physical origin, without invoking EPs. The radiative-diffraction photonic work explicitly states that its flat bands “do not rely on non-Hermitian exceptional points,” while the honeycomb wire-network work uses “exceptional” for exact, symmetry-enforced flatness across the Brillouin zone (Lehikoinen et al., 25 Feb 2026, Liu et al., 5 Feb 2026).

Setting	Meaning of “exceptional”	Representative source
Non-Hermitian EP flat band	Defective flat band at or beyond an EP	(Biesenthal et al., 2019)
NH bipartite crystal	Degeneracy-mismatch flat bands plus EP-induced spectral collapse	(Esparza et al., 14 Aug 2025)
Radiative photonics	Angle-independent diffraction flat band, explicitly not EP-based	(Lehikoinen et al., 25 Feb 2026)
Symmetry-enforced networks	Exact flatness protected by local and lattice symmetry	(Liu et al., 5 Feb 2026)

A useful distinction therefore separates flatness itself from the mechanism producing it. The surveyed literature includes destructive-interference flat bands, bipartite-rank-mismatch flat bands, non-Hermitian Aharonov–Bohm caging, moiré quasi-flat bands, and pure-diffraction flat bands. This suggests that “exceptional” functions partly as a mechanistic label and partly as a descriptor of unusual robustness or phenomenology.

2. Non-Hermitian lattice mechanisms

The most general non-Hermitian bipartite construction in the supplied literature starts from a Bloch Hamiltonian of block form

$$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$

with independent $T(k)$ and $S(k)$ in the non-Hermitian case. If sublattice $A$ hosts a momentum-independent eigenvalue $\epsilon_0$ with degeneracy $m_A$, and the inter-sublattice coupling has rank $r_k=\mathrm{rank}\,S_k$, then the flat-band counting is

$n_{\mathrm{flat},A}(k)=m_A-r_k>0.$

In the simplest fully degenerate case, $n_{\mathrm{flat},A}(k)=N_A-r_k$. This criterion is explicitly independent of Hermiticity; gain, loss, and non-reciprocity do not remove the mismatch-protected flat bands as long as the momentum-independent sublattice eigenvalue and the rank inequality are preserved (Esparza et al., 14 Aug 2025).

A particularly transparent family is the chiral non-Hermitian bipartite model

$$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$0

for which the dispersive sector is

$$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$1

At $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$2, the dispersive branches coalesce, the Hamiltonian becomes defective, and Jordan blocks appear. For $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$3, the same branches become purely imaginary; the theory interprets the resulting spectral collapse as the emergence of exceptional flat bands with tunable lifetimes,

$$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$4

The same framework identifies biorthogonal left and right eigenvectors as essential data of the flat-band problem rather than a technical afterthought (Esparza et al., 14 Aug 2025).

Several exactly solvable non-Hermitian lattice models realize this logic in concrete form. In the non-Hermitian Creutz ladder, balanced intra-leg hoppings and non-reciprocal cross-links permit both Hermitian diabolical flat bands and non-Hermitian exceptional flat bands. Under periodic boundary conditions, the sufficient EFB conditions are

$$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$5

for which $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$6 and the Bloch Hamiltonian is EP2-defective across the Brillouin zone. Under open boundary conditions the same line decomposes into order-2 Jordan blocks, while exceptional lines organize real, imaginary, and complex spectral sectors (Dutta et al., 23 Dec 2025).

A related but older photonic construction uses non-Hermitian coupling rather than on-site loss. In the diamond chain, balanced gain and loss generate an effective $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$7 flux, producing a non-Hermitian analogue of Aharonov–Bohm caging. At $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$8, one dispersive band coalesces onto the $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$9 flat band for every $T(k)$0, the Bloch Hamiltonian becomes defective for all $T(k)$1, and transport is caged. In the stub lattice, by contrast, the effective flux vanishes; the flat band remains, but it intersects a dispersive band only at isolated EPs, yielding asymmetric diffraction and anomalous linear amplification rather than full caging (Leykam et al., 2017).

3. PT-symmetric photonic realization at an exceptional point

The clearest experimental realization of an EP flat band in the provided corpus is the passive $T(k)$2-symmetric waveguide array formed by a quasi-one-dimensional tripartite lattice with sublattices $T(k)$3 and couplings $T(k)$4 and $T(k)$5. Its Bloch Hamiltonian is

$T(k)$6

which implements passive $T(k)$7 symmetry by shifting the balanced on-site set $T(k)$8 by a uniform $T(k)$9. For this system the flat-band propagation constant is

$S(k)$0

and the $S(k)$1 threshold in the balanced representation is

$S(k)$2

With $S(k)$3 and $S(k)$4, the experiment is tuned exactly to the EP (Biesenthal et al., 2019).

At that point the flat-band CLS is exceptionally simple. The paper gives the real-space compact eigenstate

$S(k)$5

a four-site “trapezoidal” pattern spanning two adjacent unit cells. Its phases enforce destructive interference both through the $S(k)$6-site couplings and across the $S(k)$7–$S(k)$8 chain coupling at the cell boundary, so the state does not radiate into the rest of the lattice. Because $S(k)$9 is $A$0-independent, arbitrary spatial superpositions of these CLS patterns preserve their transverse profile and exhibit arrested transverse dynamics (Biesenthal et al., 2019).

Experimentally, the lattice was written in fused silica by femtosecond-laser direct writing. Three distinct imaginary on-site energies $A$1 were implemented with microscopic scattering centers inserted into selected waveguides; each dot expels a small fraction of light, typically $A$2, and the total loss is set by dot spacing and dwell time. Excitation used a He–Ne laser at $A$3 and a spatial light modulator, while propagation was monitored by fluorescence microscopy around $A$4 with spectral and Fourier filtering. For the EP setting $A$5, single-site excitations populated dispersive bands and diffracted strongly, whereas the trapezoidal CLS propagated without measurable transverse broadening over the observable distance; the normalized second moment $A$6 remained essentially constant after excluding the first $A$7, where stray-light fluorescence dominates (Biesenthal et al., 2019).

This experiment established a central theme of non-Hermitian flat-band engineering: loss is not merely a perturbation to compensate, but a control parameter that can flatten bands, tune the system to an EP, and stabilize compact nondiffracting modes without additional diffraction management. The trade-off is global attenuation, since the passive implementation inherits the imaginary on-site offsets.

4. Material, network, and diffractive realizations beyond the EP paradigm

Electronic materials realize flat bands in a different register. In the weakly correlated semimetal $A$8, ARPES observes a flattened feature $A$9 around the surface $\epsilon_0$0 point of the (001) Brillouin-zone projection. Along $\epsilon_0$1–$\epsilon_0$2–$\epsilon_0$3, the dispersion is below the $\epsilon_0$4 energy resolution, implying $\epsilon_0$5 in that direction; along $\epsilon_0$6–$\epsilon_0$7–$\epsilon_0$8 the feature remains only weakly dispersive. Polarization analysis shows strong enhancement in LH and suppression in LV, consistent with predominantly Ru $\epsilon_0$9 character, while a Dirac-like surface-derived, non-topological crossing appears at $m_A$0 within the same low-energy window (Sakhya et al., 2023). In this paper the “exceptional” character is the unusual coexistence of anisotropic low-$m_A$1 flat states and Dirac-like states in a weakly correlated centrosymmetric tetragonal host rather than EP physics.

Kagome-derived intermetallics supply another materials route. In CoSn-type compounds, kagome-lattice quantum interference produces narrow bands whose usefulness depends on their proximity to $m_A$2. CoSn is singled out because the top of its low-dispersion kagome bands lies approximately at $m_A$3 and $m_A$4 relative to $m_A$5, with a distinct $m_A$6 flat band near $m_A$7; susceptibility modeling places a DOS step only $m_A$8–$m_A$9 below $r_k=\mathrm{rank}\,S_k$0. RhPb has flatter bands deeper in energy, near $r_k=\mathrm{rank}\,S_k$1, and the paper estimates that $r_k=\mathrm{rank}\,S_k$2 holes per formula unit would be required to move $r_k=\mathrm{rank}\,S_k$3 into that manifold. PtTl would require $r_k=\mathrm{rank}\,S_k$4 holes per formula unit, and NiIn exhibits broader, more three-dimensionally dispersive bands (Meier et al., 2020).

A closely related but structurally distinct $r_k=\mathrm{rank}\,S_k$5-orbital realization is provided by $r_k=\mathrm{rank}\,S_k$6 compounds. There, one near-flat band along $r_k=\mathrm{rank}\,S_k$7 originates from Co-$r_k=\mathrm{rank}\,S_k$8 and $r_k=\mathrm{rank}\,S_k$9 orbitals on the kagome plane, while an extended flat band on the $n_{\mathrm{flat},A}(k)=m_A-r_k>0.$0 plane derives from Co-$n_{\mathrm{flat},A}(k)=m_A-r_k>0.$1 and $n_{\mathrm{flat},A}(k)=m_A-r_k>0.$2 orbitals on both kagome and honeycomb planes. In both cases destructive interference among Kagome–Kagome and Kagome–honeycomb hoppings guarantees localized eigenstates and $n_{\mathrm{flat},A}(k)=m_A-r_k>0.$3-independent eigenvalues. First-principles calculations show that even $n_{\mathrm{flat},A}(k)=m_A-r_k>0.$4 changes of $n_{\mathrm{flat},A}(k)=m_A-r_k>0.$5 and $n_{\mathrm{flat},A}(k)=m_A-r_k>0.$6, or $n_{\mathrm{flat},A}(k)=m_A-r_k>0.$7 changes of $n_{\mathrm{flat},A}(k)=m_A-r_k>0.$8 at fixed $n_{\mathrm{flat},A}(k)=m_A-r_k>0.$9, preserve the near-flat dispersion while shifting the band relative to $n_{\mathrm{flat},A}(k)=N_A-r_k$0, which is why the paper frames these bands as robust and engineerable (Ochi et al., 2014).

Exact symmetry-enforced flatness also occurs in network models. Periodic honeycomb wire networks with local $n_{\mathrm{flat},A}(k)=N_A-r_k$1 vertex symmetry possess exact flat bands spanning the entire Brillouin zone for arbitrary $n_{\mathrm{flat},A}(k)=N_A-r_k$2-symmetric unitary vertex scattering matrices. The flat-band quantization is

$n_{\mathrm{flat},A}(k)=N_A-r_k$3

independent of the $n_{\mathrm{flat},A}(k)=N_A-r_k$4-phases that parameterize the symmetric versus two-dimensional irreducible sectors of the vertex $n_{\mathrm{flat},A}(k)=N_A-r_k$5-matrices. In the multichannel problem this produces a universal $n_{\mathrm{flat},A}(k)=N_A-r_k$6 ratio between flat and dispersive bands, and the flat bands persist for any number of transverse modes (Liu et al., 5 Feb 2026).

At the opposite end of the design spectrum are atomically assembled quasi-one-dimensional lattices on chlorine-vacancy arrays on Cu(100). STM/STS experiments realize diamond, cross, stub, extended-diamond, and symmetry-modified chains with flat bands placed near the single-vacancy energy $n_{\mathrm{flat},A}(k)=N_A-r_k$7. The fitted couplings are $n_{\mathrm{flat},A}(k)=N_A-r_k$8, $n_{\mathrm{flat},A}(k)=N_A-r_k$9, $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$00, and $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$01. These structures realize both gapless and gapped flat bands, as well as on/off switching of flatness by breaking and restoring mirror symmetry (Huda et al., 2020).

Two further realizations broaden the meaning of exceptional flatness. In the magic-angle vibrating-plate analogue of twisted bilayer graphene, quasi-flat flexural-wave bands arise from chirality-driven suppression of the Dirac velocity rather than destructive-interference CLS. For $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$02 and $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$03, the first magic angle is $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$04, where the Dirac-point velocity vanishes and the dimensionless Dirac frequency is $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$05, corresponding to roughly $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$06 for the proposed LiNbO$$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$07/rubber device (Rosendo et al., 2020). In radiatively coupled nanoparticle lattices built from superposed equispaced chains, flat bands instead arise from diffraction: their frequency is fixed by the chain spacing,

$$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$08

and is independent of in-plane angle. These bands are explicitly described as linearly polarized and non-EP-based; experimentally, gold nanocylinder arrays exhibit TM flat bands at $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$09 with linewidth $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$10 in a simple rectangular lattice, and at $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$11 with linewidth $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$12 and $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$13 in a rectangular chain lattice (Lehikoinen et al., 25 Feb 2026).

5. Topology, boundary accumulation, and dynamical signatures

Exceptional flat bands inherit the non-Hermitian distinction between right and left eigenvectors, so their geometric and boundary properties need not coincide with Hermitian intuition. In the general NH bipartite construction, the flat bands pinned by degeneracy mismatch have trivial Berry curvature in the chiral models discussed, but the quantum metric can acquire additional non-Hermitian divergences near high-symmetry points when spectral collapse occurs onto an exceptional flat band. The paper emphasizes that the flat-band counting is a bulk property and does not require point-gap topology or the non-Hermitian skin effect, although non-reciprocity can induce skin modes under open boundaries if spectral winding is present (Esparza et al., 14 Aug 2025).

A more radical boundary phenomenon is the “exceptional skin effect” on a flat band. In the three-site-unit-cell chiral lattice studied in (Wang et al., 2024), the flat band remains pinned at $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$14 by sublattice symmetry, and its Bloch spectrum is a single point, so the conventional point-gap winding at $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$15 is trivial:

$$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$16

Nevertheless, when nonreciprocal parameters are tuned into a gap-closed region, the flat-band Riemann sheet connects to dispersive sheets at EP lines, the generalized Brillouin-zone solution satisfies $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$17, and the flat-band eigenstates localize exponentially at a boundary. In the four-unit-cell realization, pairwise hybridization with dispersive bands generates third-order EPs; in the twelve-unit-cell elastic-wave experiment, flat-band modes localize at the right boundary throughout the gap-closed region, not only exactly at an EP (Wang et al., 2024).

The non-Hermitian Creutz ladder complements this with an analytically tractable skin-effect and bulk–boundary-correspondence problem. Its imaginary gauge transformation yields a growth/decay factor

$$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$18

whose modulus controls boundary accumulation. Along the exceptional lines $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$19 and $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$20, the similarity transform becomes singular and the open-boundary Hamiltonian contains Jordan blocks; along the bulk–boundary-restoration curves

$$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$21

the non-Hermitian skin effect disappears and the periodic- and open-boundary spectra coincide (Dutta et al., 23 Dec 2025).

Topological classification of perfectly flat bands can also be phrased without reference to EPs. The bipartite crystalline-lattice construction across all 1651 Shubnikov space groups classifies flat bands by the formal band-representation difference

$$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$22

If $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$23 at the relevant little groups, the flat band can be locally gapped; otherwise formal representation differences enforce symmetry-protected touchings with dispersive bands. In this framework gapped flat bands are natural candidates for fragile topology, while gapless flat bands arise from symmetry-enforced crossings (Călugăru et al., 2021). By contrast, the honeycomb wire-network flat bands are topologically trivial: the existence of a complete basis of CLS implies zero Chern number, and the Berry-curvature singularity at $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$24 integrates to zero over closed regions (Liu et al., 5 Feb 2026).

Dynamically, EP flat bands differ sharply from Hermitian flat bands. Near a second-order Jordan block, time evolution contains algebraic factors, $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$25, so localization may persist without oscillatory revivals. That is why the non-Hermitian Creutz ladder distinguishes diabolical flat bands, with oscillatory compacton dynamics, from exceptional flat bands, where revivals are suppressed and evolution is algebraic (Dutta et al., 23 Dec 2025). The same distinction underlies the difference between caging at an EP and ordinary Hermitian compact localization.

6. Robustness, tuning strategies, and applications

Across the literature, robustness is highly mechanism-dependent. In passive $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$26 photonics, the principal advantage is the absence of optical gain media: tailored loss tunes the band curvature and reaches the EP, while the CLS profile is protected by interference rather than external diffraction compensation. The principal cost is global attenuation, together with sensitivity of the EP condition to fabrication tolerances in $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$27, $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$28, and $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$29 (Biesenthal et al., 2019). In NH bipartite theory, mismatch-protected flat bands survive arbitrary non-Hermitian deformations that preserve the momentum-independent sublattice eigenvalue and do not raise the coupling rank enough to remove the degeneracy mismatch; the EP-induced exceptional branches, however, remain more tuning-sensitive (Esparza et al., 14 Aug 2025).

Materials papers identify several concrete control parameters. In CoSn-type kagome compounds, modest hole doping may suffice to move CoSn into the flat-band manifold because the relevant DOS step sits only $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$30–$$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$31 below $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$32, whereas RhPb requires about $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$33 holes per formula unit and PtTl requires at least $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$34 holes per formula unit (Meier et al., 2020). In $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$35, the proposed tuning knobs are strain or pressure, chemical substitution or doping, and surface engineering, motivated by the anisotropic flat segment near $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$36 and the fact that DFT had to shift $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$37 by about $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$38 to align with ARPES (Sakhya et al., 2023). In $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$39, the relative position of the nearly flat Co-$$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$40 bands can be shifted by varying lattice parameters while leaving the dispersion almost unchanged, which is why the paper explicitly describes flat-band engineering by chemical or physical pressure (Ochi et al., 2014).

Geometric design offers similarly direct control in synthetic platforms. In the radiative diffraction setting, the flat-band energy is set only by the chain spacing $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$41 and the refractive index $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$42, while the observation direction is perpendicular to the chain orientation; multi-copy offsets and oblique supercells tune angular range, polarization purity, and flatness without changing $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$43 (Lehikoinen et al., 25 Feb 2026). In the vibrating-plate moiré analogue, the twist angle $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$44 and interlayer-coupling parameter $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$45 govern the effective Bistritzer–MacDonald coupling ratio and thus the magic-angle condition (Rosendo et al., 2020). In the chlorine-vacancy atomic lattices, the decisive knobs are connectivity and symmetry: mirror breaking switches flat bands off, mirror restoration can restore or multiply them, and longer-range hoppings select which CLS survive (Huda et al., 2020).

Applications follow directly from vanishing group velocity and enhanced DOS, but their form depends on the platform. The passive $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$46 photonic experiment explicitly points to optical buffering, on-chip delay lines, memory elements, and low-footprint signal-processing stages (Biesenthal et al., 2019). The radiative flat-band metasurface work identifies slow light, nonlinear enhancement, flat-band lasing, and control of emission or absorption over a wide spectral range (Lehikoinen et al., 25 Feb 2026). In weakly correlated and kagome metals, flat-band-enhanced $$H(k)= \begin{pmatrix} H_A(k) & T(k)\ S(k) & H_B(k) \end{pmatrix},$$47 is discussed as a route to instabilities under tuning, including superconductivity or magnetic order (Sakhya et al., 2023, Meier et al., 2020). In non-Hermitian quantum crystals, tunable complex energies and lifetimes suggest open-system routes to correlated and topological phases that have no closed-system analogue (Esparza et al., 14 Aug 2025).

The combined record suggests a layered taxonomy rather than a single universal object. One branch centers on EP defectiveness, biorthogonality, and non-Hermitian spectral collapse; another on exact symmetry-enforced flatness; another on moiré chirality; and another on pure diffraction in radiative systems. What unifies them is not a single microscopic mechanism, but the recurrent conjunction of dispersion quenching, localization, and unusually strong control over transport, lifetime, or density of states.