Photonic Band Diagrams: Theory & Applications

• Photonic band diagrams are graphical representations of Bloch-mode dispersion in periodic media, mapping allowed bands and forbidden gaps.
• They utilize both direct (ω(k)) and inverse (k(ω)) formulations to address scenarios with loss, non-Hermiticity, and time-periodicity.
• Experimental reconstructions use methods like Fourier imaging spectroscopy and far-field photoluminescence to validate theoretical models.

Photonic band diagrams (BDs) are the graphical expression of the dispersion law of Bloch modes in periodic electromagnetic media. In the standard bulk setting they plot band frequencies as functions of Bloch wavevector, $\omega_n(\mathbf{k})$, over the first Brillouin zone or along a reduced symmetry path; in open, lossy, dispersive, or time-periodic systems the relevant object can instead be a complex-frequency band structure, a complex-wavevector band structure, or a related isofrequency representation. As such, BDs encode allowed bands, forbidden gaps, group velocities, evanescent decay, radiative linewidths, and, in slab and non-Hermitian settings, phenomena such as bound states in the continuum (BICs), exceptional points, and momentum gaps (Gralak et al., 2018, Rybin et al., 2017, Park et al., 2021).

1. Formal definition and canonical Bloch formulation

In the lossless periodic setting, the starting point is a periodic constitutive profile, for example $$\varepsilon(\mathbf{x}+\mathbf{R})=\varepsilon(\mathbf{x})$$ on a lattice $L$. Floquet–Bloch decomposition reduces Maxwell’s equations to a family of unit-cell eigenproblems indexed by $\mathbf{k}$ in the first Brillouin zone $B$, and the resulting eigenvalue relation defines the dispersion law $\omega=\omega_n(\mathbf{k})$ (Gralak et al., 2018). In the formulation reviewed for photonic crystals, the Bloch-reduced operator $M(\mathbf{k})$ satisfies

$$M(\mathbf{k})F_{\#}(\mathbf{x},\mathbf{k},\omega)=\omega\,F_{\#}(\mathbf{x},\mathbf{k},\omega),$$

with the usual Bloch-periodic boundary condition on opposite sides of the unit cell (Gralak et al., 2018).

For two-dimensional scalar reductions, the same paper gives the standard $s$- and $p$-polarized forms in terms of $$\varepsilon(\mathbf{x}+\mathbf{R})=\varepsilon(\mathbf{x})$$0, making explicit that the BD is the spectrum of a $$\varepsilon(\mathbf{x}+\mathbf{R})=\varepsilon(\mathbf{x})$$1-dependent Maxwell operator rather than merely a graphical convention (Gralak et al., 2018). In this Hermitian regime, the group velocity is

$$\varepsilon(\mathbf{x}+\mathbf{R})=\varepsilon(\mathbf{x})$$2

so slopes and curvatures of the bands are directly tied to transport, directional emission, self-collimation, and extreme anisotropy (Gralak et al., 2018).

A conventional high-symmetry-path plot is therefore only a reduced view of the full spectral object. This becomes especially important once absorption or material dispersion is introduced, because the spectrum is no longer confined to the real axis and the usual reduced contour can miss relevant features of the full Brillouin-zone image in the complex plane (Gralak et al., 2018).

2. Complex, inverse, and non-Hermitian band diagrams

A major generalization of the BD concept replaces the direct problem $$\varepsilon(\mathbf{x}+\mathbf{R})=\varepsilon(\mathbf{x})$$3 by the inverse problem $$\varepsilon(\mathbf{x}+\mathbf{R})=\varepsilon(\mathbf{x})$$4. For periodic media with arbitrary frequency-dependent dielectric response, including lossy media, the inverse dispersion method fixes a real frequency and solves directly for the generally complex Bloch wavevector. In the two-dimensional TE/TM plane-wave formulation, the core generalized eigenproblem is

$$\varepsilon(\mathbf{x}+\mathbf{R})=\varepsilon(\mathbf{x})$$5

which is non-Hermitian even for real constitutive functions because the eigenvalue is the wavevector magnitude $$\varepsilon(\mathbf{x}+\mathbf{R})=\varepsilon(\mathbf{x})$$6, not the frequency (Rybin et al., 2017). This formulation makes propagating branches $$\varepsilon(\mathbf{x}+\mathbf{R})=\varepsilon(\mathbf{x})$$7 and evanescent branches $$\varepsilon(\mathbf{x}+\mathbf{R})=\varepsilon(\mathbf{x})$$8 visible on the same footing, allows measured $$\varepsilon(\mathbf{x}+\mathbf{R})=\varepsilon(\mathbf{x})$$9 to be inserted directly, and exposes branch points at band edges that are hidden in direct $L$0 plots (Rybin et al., 2017).

The same conceptual move appears in three dimensions in the finite-element complex-wavenumber method for 3D metamaterial and photonic-crystal systems. There the Bloch vector is written as $L$1, and FEM discretization produces a quadratic eigenvalue problem

$L$2

so that one solves directly for complex $L$3 at fixed real $L$4 (Fietz et al., 2011). This is particularly convenient for dispersive materials such as plasmonic metals, for evanescent decay lengths, and for isofrequency contours of 3D metamaterials (Fietz et al., 2011).

Non-Hermitian complex BDs also arise when the eigenvalue remains frequency-like. In one-dimensional $L$5-symmetric photonic crystals, the complex band structure shows PT-exact regions with real eigenfrequencies and PT-broken regions where imaginary parts bifurcate after spectral singularities. The paper reports two specific transition types: re-entry from a PT-broken phase into a PT-exact phase at larger non-Hermiticity, and coalescence of two spectral singularities—one originating from the Brillouin-zone center and one from the zone boundary—into a higher-order singularity inside the Brillouin zone (Ding et al., 2015).

Time-periodic media further enlarge the notion of a BD. In a photonic Floquet medium, the experimentally relevant structures are described as measurable subsets of complex eigenfrequency surfaces over complex momentum space. The Bloch-Floquet band structure corresponds to $L$6, whereas the non-Bloch band structure is defined along the contour $L$7 in the complex Brillouin zone (Park et al., 2021). In that setting, driving-induced non-reciprocal coupling between oppositely signed frequency states opens momentum gaps rather than ordinary frequency gaps, and the gap edges are exceptional phase transitions (Park et al., 2021).

3. Numerical constructions and model classes

The numerical realization of a BD depends strongly on the physical class of periodic medium. For ordinary dielectric photonic crystals, plane-wave expansion remains a central route. In the inverse-dispersion formulation, TE and TM problems are written as plane-wave matrix eigenproblems with diagonal $L$8, $L$9 blocks and Toeplitz material matrices $\mathbf{k}$0, $\mathbf{k}$1, and convergence against the traditional $\mathbf{k}$2 plane-wave method was reported with $\mathbf{k}$3 plane waves for the two-dimensional square lattices studied (Rybin et al., 2017).

For open atomic lattices, the band structure is instead built from a non-Hermitian Bloch Hamiltonian derived from the electromagnetic dyadic Green’s function. In an infinite two-dimensional atomic array, the Bloch eigenvalues take the form

$\mathbf{k}$4

so the real part gives the collective frequency shift while the imaginary part gives the collective radiative linewidth (Perczel et al., 2017). The computational difficulty of long-range $\mathbf{k}$5 dipole coupling is handled by converting real-space lattice sums to reciprocal-space sums with Poisson summation and Weyl decomposition, yielding rapidly convergent complex band structures for Bravais and non-Bravais lattices alike (Perczel et al., 2017).

In plasma photonic crystals, the dispersion can be extracted kinetically rather than from a direct eigensolver. A one-dimensional Particle-in-Cell simulation excites the structure with a broadband pulse,

$\mathbf{k}$6

records $\mathbf{k}$7, and obtains the dispersion by a two-dimensional Fourier transform in space and time (Trieschmann et al., 2017). The resulting $\mathbf{k}$8 data are then reformatted into a BD representation in terms of

$\mathbf{k}$9

showing that fully kinetic simulations of plasma–dielectric multilayers agree well with the cold-plasma description while preserving the usual photonic-crystal language of bands and gaps (Trieschmann et al., 2017).

One-dimensional non-Hermitian periodic media admit yet another route. In $B$0-symmetric layered crystals, the paper constructs a Hamiltonian in the basis of Bloch states of the passive crystal, so that the complex BD is obtained by diagonalizing a finite-dimensional non-Hermitian matrix at each $B$1 rather than by solving the full complex-wave equation from scratch (Ding et al., 2015). This suggests that, across very different platforms, the core object is still a Bloch-reduced operator, but its natural basis may range from plane waves to Green-function-mediated dipole states to Hermitian passive-crystal bands.

4. Experimental reconstruction of band diagrams

Experimental BDs are commonly reconstructed from angle- or momentum-resolved observables rather than measured as direct eigenvalue spectra. A clear example is the wide-angle near-infrared Fourier imaging spectroscopy study of a three-dimensional rod-connected-diamond photonic crystal. There the measured object is an angularly and spectrally resolved reflectance map, later associated with the irreducible Brillouin-zone path $B$2. The resulting contour plot of unpolarized reflectivity versus wavelength and reciprocal-space path shows the fundamental gap between the 2nd and 3rd bands, with a center wavelength shifting from about $B$3 to $B$4 and a width of roughly $B$5, in approximate agreement with plane-wave and FDTD calculations (Chen et al., 2017).

A different strategy uses the structure itself as an internal light source. In freestanding SiN$B$6 photonic crystal slabs, momentum-resolved far-field photoluminescence reproduces the energy–momentum dispersions of TE-like and TM-like slab bands, and a simple bandpass-filter modification yields direct isofrequency contours in the back focal plane (Lan et al., 2023). In that platform, symmetry-protected $B$7-point BICs appear as dark points on otherwise bright bands, and polarization-resolved Stokes maps reveal an azimuthally polarized vortex with topological charge

$B$8

around the BIC (Lan et al., 2023).

A later slab experiment extends this logic from an isolated quasi-BIC point to a quasi-bound band in the continuum. Wavevector-resolved photoluminescence of monolayer WSe$B$9 weakly coupled to a square lattice of aluminum nanodisks shows that the symmetry-protected quasi-BIC at $\omega=\omega_n(\mathbf{k})$0 is the endpoint of a narrow upper band extending along the $\omega=\omega_n(\mathbf{k})$1-X direction. The measured linewidth of this qBBC remains in the $\omega=\omega_n(\mathbf{k})$2–$\omega=\omega_n(\mathbf{k})$3 range through at least half of the Brillouin zone, while the bright low-energy edge of the stop band near $\omega=\omega_n(\mathbf{k})$4 is broader, with FWHM $\omega=\omega_n(\mathbf{k})$5 (Tsoi et al., 5 Nov 2025). This suggests that experimental BDs in open photonic slabs increasingly function as combined dispersion–linewidth maps rather than purely real-frequency plots.

5. Reduced representations, classification, and surrogate generation

Although the standard BD is branch-resolved in $\omega=\omega_n(\mathbf{k})$6-space, several works recast the same information into more design-oriented representations. One example is the photonic density of states map, defined from the full Brillouin-zone band structure as

$\omega=\omega_n(\mathbf{k})$7

In two-dimensional rod lattices, the resulting PhDOS map is plotted versus normalized frequency and rod-radius ratio $\omega=\omega_n(\mathbf{k})$8, and was reported to show full correspondence with independently computed transmittance maps while requiring substantially less computation time than high-resolution FDTD transmittance sweeps (Sukhoivanov et al., 2010). In that representation, zero DOS marks full photonic band gaps, low DOS marks weakly allowed regions, and high DOS tracks flat or densely populated band regions (Sukhoivanov et al., 2010).

A complementary reduction is the one-dimensional band-gap atlas for binary, ternary, and linearly graded multilayers. Rather than plotting full $\omega=\omega_n(\mathbf{k})$9, that work maps gap size $M(\mathbf{k})$0, gap position $M(\mathbf{k})$1, and normalized width $M(\mathbf{k})$2 over structural parameter space, and reports that in binary systems the $M(\mathbf{k})$3-th gap exhibits $M(\mathbf{k})$4 hills and $M(\mathbf{k})$5 valleys in the $M(\mathbf{k})$6 atlas (Pardo et al., 2024). This suggests that, for many 1D design tasks, the decisive content of a BD can be compressed into a structured parameter atlas of forbidden intervals without explicit branch-by-branch plotting.

Persistent homology pushes this compression in a different direction by classifying the topology of isoenergy sets and Bloch-eigenstate manifolds. For sampled photonic band structures $M(\mathbf{k})$7, the method studies sublevel sets of

$M(\mathbf{k})$8

and tracks Betti numbers $M(\mathbf{k})$9 and $$M(\mathbf{k})F_{\#}(\mathbf{x},\mathbf{k},\omega)=\omega\,F_{\#}(\mathbf{x},\mathbf{k},\omega),$$0 to distinguish isolated valleys, moat bands, and multiply connected low-energy contours (Leykam et al., 2020). The same paper shows that a quantum-distance graph of Bloch eigenstates,

$$M(\mathbf{k})F_{\#}(\mathbf{x},\mathbf{k},\omega)=\omega\,F_{\#}(\mathbf{x},\mathbf{k},\omega),$$1

provides a second, state-space topology that can distinguish energy-disconnected but polarization-similar valleys from energy-disconnected and polarization-distinct valleys (Leykam et al., 2020).

More recently, BDs have also been treated as image-generation targets. A conditional transformer–latent-diffusion model was trained to map layered three-dimensional photonic-structure descriptions ({(\epsilon_1,d_1),\dots,(\epsilon