Nonlocal Effective Medium

• Nonlocal effective medium is a homogenized description of structured materials where the macroscopic response depends on both frequency and spatial variation.
• It employs derivative expansion and multiscale homogenization to capture first- and second-order spatial dispersion beyond local approximations.
• This framework elucidates phenomena such as modal splitting, angle-dependent reflectance, and modified light–matter interactions in metamaterials.

A nonlocal effective medium is a homogenized description of a structured material in which the macroscopic constitutive response depends not only on frequency but also on spatial variation, equivalently on wavevector. In electromagnetism this is expressed by replacing the local constitutive law $$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{k},\omega)$$ with a wavevector-dependent tensor $$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega)$$, or, in real space, by a convolution kernel relating fields over a finite region. Across metamaterials, multilayers, wire media, and related composite systems, nonlocal effective-medium theory is used when local homogenization reproduces neither bulk modal structure nor finite-slab observables such as reflection, transmission, or spontaneous-emission rates (Ginzburg et al., 2016, Ciattoni et al., 2015, Mnasri et al., 2018).

1. Constitutive structure and formal definition

In the local approximation, a composite is replaced by a homogeneous medium whose response at a point depends only on the field at that same point. A standard form is

$$\mathbf{D}(\mathbf{r},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{r},\omega), \qquad \mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{k},\omega).$$

A nonlocal medium instead obeys

$$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega), \qquad \mathbf{D}(\mathbf{r},\omega)=\int \boldsymbol{\epsilon}(\mathbf{r}-\mathbf{r}',\omega)\mathbf{E}(\mathbf{r}',\omega)\,d^3r',$$

so the constitutive tensor becomes wavevector dependent (Ginzburg et al., 2016).

For periodic metamaterials in the long-wavelength limit, spatial dispersion may be organized as a derivative expansion,

$$D_i(\mathbf{r})=\varepsilon_0\left[\varepsilon^{(0)}_{ij}E_j(\mathbf{r}) +\varepsilon^{(1)}_{ijr}\frac{\partial E_j}{\partial x_r} +\varepsilon^{(2)}_{ijrs}\frac{\partial^2E_j}{\partial x_r\partial x_s} +\dots\right].$$

Within this hierarchy, the zeroth-order tensor is the ordinary effective permittivity, the first-order tensor encodes first-order spatial dispersion, and the second-order tensor encodes genuine second-order nonlocality. In the multiscale homogenization of periodic non-magnetic metamaterials, first-order spatial dispersion is equivalent to a reciprocal bianisotropic response, while second-order terms remain as explicit spatial-dispersion corrections (Ciattoni et al., 2015).

A closely related formulation appears in the strong-spatial-dispersion model used for parameter retrieval from slab scattering. There the constitutive operator is written schematically as a local permittivity term plus higher-order operators, with weak spatial dispersion identified by $\gamma=0$ and strong spatial dispersion by $\gamma\neq 0$. This formulation makes explicit that a local $\varepsilon$–$\mu$ description is only a special case of a more general nonlocal constitutive law (Mnasri et al., 2018).

An important limitation follows immediately: while first-order nonlocality can often be rewritten as local bianisotropy, a local bianisotropic model does not capture all second-order nonlocal effects in multilayered metamaterials. In that sense, nonlocal effective-medium theory is not merely a more elaborate parameterization of local anisotropy; it introduces qualitatively new bulk and boundary structures (Gorlach et al., 2019).

2. Homogenization, retrieval, and analytical frameworks

Several distinct derivational routes lead to nonlocal effective media. In multiscale homogenization, rapidly varying cell-scale fields are separated from slowly varying macroscopic fields, and the effective tensors are obtained by averaging solutions of electrostatic-like cell problems. This yields explicit formulas for the local term and for the first- and second-order spatial-dispersion tensors, together with symmetry-based statements about reciprocal chirality and pseudo-chiral-omega response (Ciattoni et al., 2015).

A different route is two-scale renormalization, developed for finite-conductivity wire metamaterials. There the lattice period and wire radius are taken to zero with suitable renormalized parameters held fixed, producing a macroscopic system in which the polarization component along the wires satisfies an inhomogeneous Helmholtz equation with Neumann boundary conditions. Solving that equation gives an integral constitutive relation

$$P_z(x,z_0)=-2\pi\gamma\varepsilon_0\int_{-L/2}^{L/2} g(z,z_0)\,E_z(x,z)\,dz,$$

which is explicitly nonlocal along the wire direction (Cabuz et al., 2010).

For layered structures, transfer-matrix homogenization provides a direct route to $$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega)$$0-dependent effective tensors. In symmetric metal–dielectric multilayers under TM illumination, averaging the exact fields over a symmetric unit cell yields nonlocal effective permittivities $$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega)$$1 and $$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega)$$2. This construction retains the dependence on incidence angle through $$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega)$$3, and thereby incorporates structural spatial dispersion that is inaccessible to simple mixing formulas (Sun et al., 2015).

A complementary strategy is inverse retrieval from reflection and transmission. In the strong-spatial-dispersion approach, the slab is treated as a homogeneous nonlocal medium and the effective parameters are inferred by fitting its Fresnel response. The stated motivation is that local retrievals often fail when the incidence angle changes, and can generate anti-Lorentz resonances or negative imaginary parts of $$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega)$$4 or $$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega)$$5. The nonlocal retrieval instead keeps the higher-order constitutive structure and produces a markedly better description of oblique-incidence optics in fishnet metamaterials (Mnasri et al., 2018).

3. Canonical realizations

Nonlocal effective-medium behavior is particularly clear in a small set of canonical metamaterial platforms.

Platform	Nonlocal quantity	Characteristic consequence
Plasmonic nanorod assembly	$$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega)$$6	Additional TM branch, including hyperbolic-like behavior in a locally elliptic regime
Symmetric metal–dielectric multilayer	$$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega)$$7	Angle-dependent reflectance and bulk plasmon-polariton branches
Wire medium	$$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega)$$8, or $$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega)$$9	Persistent spatial dispersion even for very large wavelength-to-period ratio

In metallic nanorod metamaterials, the local Maxwell–Garnett picture gives a uniaxial tensor with $$\mathbf{D}(\mathbf{r},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{r},\omega), \qquad \mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{k},\omega).$$0 and $$\mathbf{D}(\mathbf{r},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{r},\omega), \qquad \mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{k},\omega).$$1, but the nonlocal theory following Wells, Zayats, and Podolskiy replaces the axial response by a rational function $$\mathbf{D}(\mathbf{r},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{r},\omega), \qquad \mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{k},\omega).$$2 with poles and zeros. The resulting TM dispersion has two branches rather than one: an elliptic-like branch and an additional hyperbolic-like branch, the latter persisting even when the local effective medium predicts an elliptic regime (Ginzburg et al., 2016, Geng et al., 2015).

In symmetric metal–dielectric multilayers, nonlocality is structural rather than hydrodynamic in origin: the field varies strongly across the unit cell, and under TM excitation the effective response depends explicitly on $$\mathbf{D}(\mathbf{r},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{r},\omega), \qquad \mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{k},\omega).$$3. The transfer-matrix-derived nonlocal tensors reproduce angle-dependent reflectance and the dispersion of bulk plasmon polaritons, whereas local effective-medium formulas shift these features incorrectly (Sun et al., 2015).

Wire media exhibit perhaps the strongest statement of principle: defining a local refractive index requires the wavelength to be large compared with the period, but the converse does not hold. Infinite wire media remain nonlocal for arbitrarily large wavelength-to-period ratios, with an effective axial permittivity of the form

$$\mathbf{D}(\mathbf{r},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{r},\omega), \qquad \mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{k},\omega).$$4

in the idealized limit. For bounded wire media, one may re-express this behavior by a local, thickness-dependent permittivity that averages the nonlocal response together with the effects of terminations and loads (Cabuz et al., 2010, Yakovlev et al., 2019).

4. Interfaces, additional waves, and boundary conditions

A decisive feature of nonlocal effective media is that the bulk constitutive law is no longer sufficient to solve finite-boundary-value problems. Because the dispersion relation can yield multiple $$\mathbf{D}(\mathbf{r},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{r},\omega), \qquad \mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{k},\omega).$$5 solutions for fixed $$\mathbf{D}(\mathbf{r},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{r},\omega), \qquad \mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{k},\omega).$$6, additional boundary conditions or equivalent interface formalisms are required.

In the strong-spatial-dispersion slab model, each polarization supports two bulk modes inside the nonlocal slab, and matching fields across the interfaces leads to a $$\mathbf{D}(\mathbf{r},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{r},\omega), \qquad \mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{k},\omega).$$7 Fresnel matrix rather than the usual $$\mathbf{D}(\mathbf{r},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{r},\omega), \qquad \mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{k},\omega).$$8 local-system algebra. This is the finite-slab manifestation of the fact that nonlocal media support more internal degrees of freedom than local ones (Mnasri et al., 2018).

In hydrodynamic models of metals, nonlocality introduces longitudinal electron-density waves in addition to transverse electromagnetic waves. The metal then requires an additional boundary condition beyond tangential $$\mathbf{D}(\mathbf{r},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{r},\omega), \qquad \mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\omega)\mathbf{E}(\mathbf{k},\omega).$$9 and $$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega), \qquad \mathbf{D}(\mathbf{r},\omega)=\int \boldsymbol{\epsilon}(\mathbf{r}-\mathbf{r}',\omega)\mathbf{E}(\mathbf{r}',\omega)\,d^3r',$$0: the normal component of the free-electron polarization is set to zero at the surface. This condition fixes the longitudinal contribution and strongly affects ultrathin metallic layers and gap-plasmon resonances (Benedicto et al., 2014).

Multilayers provide a distinct but related interface phenomenon. For deeply subwavelength all-dielectric stacks, weak nonlocality can accumulate into strong boundary effects that are not predicted by conventional mixing formulas. The mechanism can be written in terms of trace and anti-trace maps of the optical transfer matrix, so that very small unit-cell errors propagate into order-unity deviations in reflection and transmission under critical conditions (Castaldi et al., 2017).

A more general averaged-field treatment of multilayers shows that bulk nonlocal response is independent of unit-cell choice, but the effective boundary conditions are strongly sensitive to the sequence of layers and to multilayer termination. This distinction—bulk invariance with termination-dependent boundary data—is one of the clearest signatures of nonlocal homogenization (Gorlach et al., 2019).

For bounded wire media, the same issue can be recast constructively: a closed-form local thickness-dependent permittivity can be derived for structures with lumped impedance insertions and impedance-surface terminations, thereby packaging nonlocality and boundary effects into a reduced local model suitable for standard full-wave solvers (Yakovlev et al., 2019).

5. Dispersion topology and observable consequences

The most visible consequences of nonlocal effective-media theory are modal splitting, topology changes in isofrequency surfaces, altered refraction, and modified light–matter interaction. In a plasmonic nanorod assembly, the local effective-medium model yields elliptic dispersion for $$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega), \qquad \mathbf{D}(\mathbf{r},\omega)=\int \boldsymbol{\epsilon}(\mathbf{r}-\mathbf{r}',\omega)\mathbf{E}(\mathbf{r}',\omega)\,d^3r',$$1 nm, epsilon-near-zero behavior around $$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega), \qquad \mathbf{D}(\mathbf{r},\omega)=\int \boldsymbol{\epsilon}(\mathbf{r}-\mathbf{r}',\omega)\mathbf{E}(\mathbf{r}',\omega)\,d^3r',$$2 nm, and hyperbolic dispersion for $$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega), \qquad \mathbf{D}(\mathbf{r},\omega)=\int \boldsymbol{\epsilon}(\mathbf{r}-\mathbf{r}',\omega)\mathbf{E}(\mathbf{r}',\omega)\,d^3r',$$3 nm. The nonlocal model changes this picture by introducing a second TM branch, so that even the locally elliptic regime supports an additional hyperbolic-like mode. In the same class of materials, an analytical nonlocal effective-medium approximation based on Mie scattering predicts a coexistence state of two modes around the ENZ region, with both positive and negative refraction present; outside that coexistence region only one mode is excited and the local theory becomes adequate again. The same nanorod platform also shows that spontaneous emission is governed not only by the local optical topology but by the nonlocal response of the composite: metamaterials with identical local effective permittivity can exhibit different Purcell factors, a record-high enhancement of a decay rate was observed, and reducing the unit-cell scale while preserving the local tensor can yield an order-of-magnitude further enhancement of decay rates (Ginzburg et al., 2016, Geng et al., 2015).

In metal–dielectric multilayers, nonlocal effective-medium theory accurately predicts measured angle-dependent reflection spectra and the dispersion of bulk plasmon polaritons with large wavevectors, while local EMT gives pronounced spectral shifts. In hydrodynamic calculations of metallo-dielectric multilayers, nonlocality generally increases transmission through negative-index, hyperbolic, and near-zero-index structures, and it becomes especially important for gap-plasmon resonances, where confinement drives the system into the regime of large field gradients and strong longitudinal-electron response (Sun et al., 2015, Benedicto et al., 2014).

6. Misconceptions, limitations, and extensions

A persistent misconception is that subwavelength structuring automatically guarantees locality. The supplied literature makes the opposite point. Wire media remain spatially dispersive even when the wavelength is arbitrarily large compared with the lattice period, and deeply subwavelength dielectric multilayers can display strong boundary anomalies generated by weak nonlocality (Cabuz et al., 2010, Castaldi et al., 2017).

A second misconception is that nonlocality can always be absorbed into a local anisotropic or bianisotropic model. First-order spatial dispersion may indeed be recast as reciprocal bianisotropy, but local bianisotropic models do not capture all second-order nonlocal bulk effects in multilayers, and local $$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega), \qquad \mathbf{D}(\mathbf{r},\omega)=\int \boldsymbol{\epsilon}(\mathbf{r}-\mathbf{r}',\omega)\mathbf{E}(\mathbf{r}',\omega)\,d^3r',$$4–$$\mathbf{D}(\mathbf{k},\omega)=\boldsymbol{\epsilon}(\mathbf{k},\omega)\mathbf{E}(\mathbf{k},\omega), \qquad \mathbf{D}(\mathbf{r},\omega)=\int \boldsymbol{\epsilon}(\mathbf{r}-\mathbf{r}',\omega)\mathbf{E}(\mathbf{r}',\omega)\,d^3r',$$5 retrievals frequently lose predictive power away from the incidence conditions used for fitting (Ciattoni et al., 2015, Mnasri et al., 2018, Gorlach et al., 2019).

The concept also extends beyond the optical metamaterials that dominate the field. In nonlocal electrostatics, a wavevector-dependent dielectric function or an equivalent two-field formulation introduces a polarization correlation length and a polarization screening length, so that surface polarization can compete with or even invert the far-field potential generated by free charge (Behjatian et al., 2024). In effective-medium theory for Van-Der-Waals heterostructures, a locally non-periodic distribution of high-index dielectric nanoparticles is first reduced to a Foldy–Lax point-interaction system and then to a continuous Lippmann–Schwinger equation, yielding an effective medium in which the permittivity remains unchanged while the permeability is altered (Cao et al., 2024). This suggests that the defining feature of a nonlocal effective medium is not a particular material platform, but a homogenized description in which microstructural interactions survive as wavevector dependence, convolution kernels, additional waves, or termination-sensitive boundary data.