Nonlocal Temporal Effective Medium Theory

• Nonlocal Temporal EMT is a framework that replaces time-varying, memory-bearing media with effective operators that depend on frequency and wavevector.
• It simultaneously treats temporal modulation and intrinsic dispersion by averaging two-time kernels into homogenized constitutive laws with spatial nonlocal corrections.
• Implementations in acoustics, electromagnetics, and metasurfaces reveal practical averaging rules, boundary effects, and challenges for extending the theory to fully spatiotemporal systems.

Searching arXiv for papers on nonlocal temporal effective medium theory and closely related temporal/spatiotemporal homogenization. Nonlocal temporal effective medium theory is the homogenization framework for media whose constitutive response varies rapidly in time while retaining memory, so that the coarse-grained medium is described by effective parameters that remain frequency-dependent and may also acquire wavevector dependence. In this setting, temporal modulation and intrinsic dispersion are treated simultaneously rather than separately: the microscopic constitutive law is generally a two-time kernel, while the effective medium replaces that nonstationary kernel by averaged constitutive operators valid on slow time scales. Recent work has established this program for acoustics, electromagnetics, metasurfaces, and transmission-line metamaterials, showing that the effective description may take the form of averaged susceptibilities, inverse-susceptibility averages, higher-order derivative corrections, or explicit spatiotemporally dispersive operators, depending on the hierarchy among signal, resonance, and modulation frequencies (Zhu et al., 9 May 2025, Döding et al., 27 May 2025, Rizza et al., 2022, Wen et al., 2021, Overvig et al., 2023, Ciabattoni et al., 2 Jan 2026).

1. Definition and conceptual scope

Nonlocal temporal effective medium theory extends effective medium theory from static or spatially periodic media to systems whose constitutive parameters are periodic, switched, or otherwise rapidly varying in time. In the most direct formulation, the goal is to replace a time-dependent microscopic medium by an effective homogeneous medium that reproduces the low-frequency dynamics averaged over many modulation periods (Zhu et al., 9 May 2025).

A central distinction is between two kinds of nonlocality. First, temporal nonlocality means memory: the response at time $t$ depends on earlier times $t'$, typically through a constitutive kernel such as

$M(t) = \int_{-\infty}^t \chi(t,t')\,p(t')\,dt'.$

Second, spatial nonlocality means spatial dispersion: the effective constitutive parameters depend on wavevector $\mathbf{k}$, or equivalently on spatial derivatives of the fields. Several recent works show that rapid temporal modulation can induce spatially nonlocal corrections even when the microscopic material is spatially homogeneous (Rizza et al., 2022), while others retain temporal nonlocality at the coarse-grained level through effective dispersive susceptibilities (Zhu et al., 9 May 2025, Wen et al., 2021).

The theory is therefore not a single formula but a family of homogenization procedures. In dispersive acoustic metamaterials with resonant microdynamics, the effective medium remains frequency-dependent after temporal averaging, because the underlying resonances store memory (Zhu et al., 9 May 2025). In purely time-varying dielectric media treated by asymptotic homogenization, leading-order constitutive laws are local, whereas second-order corrections introduce nonlocal constitutive relations involving higher-order operators (Döding et al., 27 May 2025). In spatiotemporally modulated photonic or transmission-line systems, the effective medium can become both non-Hermitian and spatially dispersive, with generalized Brillouin-zone structure and engineered $k$-dependent dispersion (Ding et al., 2024, Ciabattoni et al., 2 Jan 2026).

This suggests a useful unifying characterization: nonlocal temporal EMT is the effective description of rapidly time-structured media in which the homogenized constitutive law retains either memory, spatial dispersion induced by temporal modulation, or both.

2. Constitutive kernels, memory, and homogenized operators

The most general formulation arises from constitutive operators that are nonlocal in both space and time. In the effective-Hamiltonian approach to electromagnetic and quantum metamaterials, the macroscopic state $\Psi(\mathbf{r},t)$ evolves under an effective operator $\hat H_\mathrm{eff}$ acting as a space-time convolution,

$$(\hat H_\mathrm{eff}\Psi)(\mathbf r,t) = \int d^N r' \int_{0}^{t} dt'\; \mathbf h_\mathrm{eff}(\mathbf r-\mathbf r',t-t')\,\Psi(\mathbf r',t'),$$

so temporal nonlocality appears as $\omega$-dependence and spatial nonlocality as $\mathbf{k}$-dependence in $t'$0 (Silveirinha, 2016). In electromagnetism this becomes a constitutive law of the form

$t'$1

which is the canonical nonlocal effective-medium structure (Silveirinha, 2016).

A more explicit temporal-kernel construction appears in programmable acoustic metamaterials. There, each resonant branch obeys a Lorentz-type oscillator equation,

$t'$2

so each branch contributes a dispersive susceptibility $t'$3, and the full compressibility is

$t'$4

with $t'$5 (Zhu et al., 9 May 2025). Because the oscillator dynamics itself is retarded, the response is temporally nonlocal even before any modulation is introduced. Temporal modulation then breaks time-translation invariance, so the kernel becomes a genuine two-time object $t'$6 rather than a function of $t'$7 only (Zhu et al., 9 May 2025).

Digital acoustic implementations make this kernel structure literal. In dynamically switched acoustic metamaterials, each atom is programmed through a time-varying convolution kernel $t'$8 linking detector and actuator,

$t'$9

so dispersion is carried by the lag variable $M(t) = \int_{-\infty}^t \chi(t,t')\,p(t')\,dt'.$0 and modulation enters through the explicit time dependence $M(t) = \int_{-\infty}^t \chi(t,t')\,p(t')\,dt'.$1 (Wen et al., 2021). This is a direct realization of a nonstationary memory kernel.

In metasurface theory, the same principle appears in spatially extended modal form. Spatio-temporal coupled mode theory promotes the resonant amplitude $M(t) = \int_{-\infty}^t \chi(t,t')\,p(t')\,dt'.$2 to a field $M(t) = \int_{-\infty}^t \chi(t,t')\,p(t')\,dt'.$3, leading to

$M(t) = \int_{-\infty}^t \chi(t,t')\,p(t')\,dt'.$4

with nonlocal coupling kernels $M(t) = \int_{-\infty}^t \chi(t,t')\,p(t')\,dt'.$5 and $M(t) = \int_{-\infty}^t \chi(t,t')\,p(t')\,dt'.$6 (Overvig et al., 2023). Here the effective surface response is encoded by a scattering kernel $M(t) = \int_{-\infty}^t \chi(t,t')\,p(t')\,dt'.$7, which functions as a nonlocal, frequency-dependent surface operator (Overvig et al., 2023).

Across these examples, the homogenized object is not merely a scalar parameter but an effective operator. What changes from system to system is the approximation used to compress the microscopic operator into a tractable coarse-grained law.

3. Temporal averaging rules and scale hierarchies

A defining result of recent nonlocal temporal EMT is that different modulation protocols lead to different averaging rules. These rules are controlled by the hierarchy among the signal frequency $M(t) = \int_{-\infty}^t \chi(t,t')\,p(t')\,dt'.$8, the modulation frequency $M(t) = \int_{-\infty}^t \chi(t,t')\,p(t')\,dt'.$9, and the resonance frequencies $\mathbf{k}$0.

In multi-resonant acoustic metamaterials, the dynamics are recast as a first-order system

$\mathbf{k}$1

for a state vector such as

$\mathbf{k}$2

and homogenization proceeds by expanding the one-period transfer matrix under the assumption $\mathbf{k}$3 (Zhu et al., 9 May 2025). When all resonances are slow relative to the modulation, $\mathbf{k}$4, the effective generator is the temporal average

$\mathbf{k}$5

leading to explicit effective susceptibilities (Zhu et al., 9 May 2025).

The resulting averaging rules differ according to what is modulated. If the resonant strength $\mathbf{k}$6 is modulated while $\mathbf{k}$7 and $\mathbf{k}$8 are fixed, then each resonant susceptibility obeys the arithmetic average

$\mathbf{k}$9

If instead the resonant frequency $k$0 is modulated while $k$1 and $k$2 are fixed, then the effective rule is reciprocal averaging,

$k$3

These are not interchangeable: they reflect different microscopic ways in which modulation enters the oscillator equation (Zhu et al., 9 May 2025).

A third regime occurs when resonances are fast relative to modulation, $k$4. Then the resonant response is effectively instantaneous on the modulation timescale, and the medium reduces to an Engheta-type non-dispersive temporal EMT with

$k$5

In hybrid cases, a fast resonance can be absorbed into a renormalized non-dispersive background while slow resonances are homogenized relative to that background (Zhu et al., 9 May 2025).

The same theme appears in programmable acoustic experiments, where the effective compressibility of a medium switching between two configurations obeys

$k$6

provided the modulation frequency is much larger than the signal frequency (Wen et al., 2021). The same temporal averaging applies to density and Willis coupling parameters when the full constitutive matrix is averaged over one modulation period (Wen et al., 2021).

A key point, emphasized in both analytical and experimental work, is that the averaged quantity depends on the microscopic implementation. In the programmable acoustic platform, modulating the numerator of the digital filter corresponds to averaging compressibility $k$7, whereas other formulations may average bulk modulus instead (Wen et al., 2021). This is a source of recurring confusion in the literature and is not a contradiction: the correct averaging rule is representation-dependent.

4. Emergent spatial dispersion and higher-order homogenization

A major development in nonlocal temporal EMT is the recognition that rapid time modulation can generate effective spatial nonlocality, even when the microscopic medium has no intrinsic spatial microstructure.

For temporally periodic dielectric media with $k$8, a multiscale expansion in $k$9 yields effective Maxwell equations for the averaged fields,

$\Psi(\mathbf{r},t)$0

together with a constitutive relation

$\Psi(\mathbf{r},t)$1

(Rizza et al., 2022). In the spatially homogeneous case this reduces to

$\Psi(\mathbf{r},t)$2

which is explicitly spatially nonlocal because of the Laplacian term (Rizza et al., 2022).

In Fourier space, the same result can be interpreted as a $\Psi(\mathbf{r},t)$3-dependent effective permeability,

$\Psi(\mathbf{r},t)$4

so a purely dielectric time-modulated medium acquires effective nonlocal magnetism (Rizza et al., 2022). Under positive permittivity modulation this response is diamagnetic, with $\Psi(\mathbf{r},t)$5, and no $\Psi(\mathbf{r},t)$6-negative or paramagnetic behavior appears within the stated approximation (Rizza et al., 2022).

A mathematically parallel result was later obtained through formal two-scale homogenization of time-varying media. For the electric case

$\Psi(\mathbf{r},t)$7

the leading homogenized equation is local,

$\Psi(\mathbf{r},t)$8

but the second-order macroscopic approximation satisfies

$\Psi(\mathbf{r},t)$9

(Döding et al., 27 May 2025). Recast in Maxwell form, this becomes

$\hat H_\mathrm{eff}$0

which is a nonlocal constitutive relation (Döding et al., 27 May 2025).

These results are closely related but differ in emphasis. The earlier multiscale electromagnetic theory directly highlights emergent $\hat H_\mathrm{eff}$1-dependent magnetism and nonlocal temporal boundary conditions (Rizza et al., 2022). The later formal homogenization analysis stresses the order at which nonlocality first appears: leading and first-order laws are local, while second-order corrections generate the nonlocal constitutive operator (Döding et al., 27 May 2025). Taken together, they establish that nonlocal temporal EMT may generate spatial-dispersive corrections even in media modulated only in time.

5. Interfaces, boundaries, and finite-size effects

Temporal homogenization is not determined solely by bulk dispersion; boundary conditions can also become nonlocal. This is particularly explicit for temporal boundaries, where a static medium is abruptly switched into a temporally periodic medium. In that setting, the averaged spectral amplitudes satisfy jump conditions

$\hat H_\mathrm{eff}$2

$\hat H_\mathrm{eff}$3

with nonlocal boundary parameters

$\hat H_\mathrm{eff}$4

(Rizza et al., 2022). These boundary terms vanish in the $\hat H_\mathrm{eff}$5 limit but can be substantial at finite $\hat H_\mathrm{eff}$6, so the effective theory must include interface corrections, not only bulk parameters (Rizza et al., 2022).

This nonlocal temporal boundary theory supports a specific wave-processing functionality. Under impedance matching, $\hat H_\mathrm{eff}$7, the reflected electric field from a temporal boundary acquires a response

$\hat H_\mathrm{eff}$8

so the reflected field becomes proportional to the first spatial derivative of the incident wavepacket (Rizza et al., 2022). Local EMT misses this effect entirely because it predicts zero reflection at impedance matching (Rizza et al., 2022).

Boundary sensitivity also has an instructive spatial analog in weakly nonlocal multilayers. There, small unit-cell nonlocal errors in the trace and anti-trace of the transfer matrix can accumulate into large finite-slab transmission errors, even when bulk dispersion is accurately reproduced by local EMT (Castaldi et al., 2017). The exact scaling laws are

$\hat H_\mathrm{eff}$9

$$(\hat H_\mathrm{eff}\Psi)(\mathbf r,t) = \int d^N r' \int_{0}^{t} dt'\; \mathbf h_\mathrm{eff}(\mathbf r-\mathbf r',t-t')\,\Psi(\mathbf r',t'),$$0

with strong boundary effects emerging after many unit cells through transfer-matrix iteration (Castaldi et al., 2017). Although this work is spatial rather than temporal, it clarifies a general lesson relevant to temporal EMT: matching bulk dispersion alone does not guarantee accurate finite-system behavior (Castaldi et al., 2017).

This suggests a broader principle. In nonlocal temporal EMT, especially for time-switched or temporally layered systems, the effective description should be judged by its ability to match transfer operators or interface maps, not only homogenized bulk spectra.

6. Platforms, applications, and related formalisms

Recent work has produced several concrete realizations of nonlocal temporal effective media.

In acoustics, programmable resonant media with multiple Lorentz branches provide an analytically tractable bulk platform in which temporal averaging rules can be derived mode by mode (Zhu et al., 9 May 2025). Digital acoustic metamaterials implement the underlying time-varying kernel directly and experimentally realize temporally averaged compressibility, density, and Willis coupling, with modulation-phase disorder found negligible in the effective regime (Wen et al., 2021).

In photonics, spatio-temporal coupled mode theory provides an effective surface theory for nonlocal metasurfaces by combining first-order temporal dynamics with lateral modal dispersion and nonlocal coupling kernels (Overvig et al., 2023). The corresponding Green’s function

$$(\hat H_\mathrm{eff}\Psi)(\mathbf r,t) = \int d^N r' \int_{0}^{t} dt'\; \mathbf h_\mathrm{eff}(\mathbf r-\mathbf r',t-t')\,\Psi(\mathbf r',t'),$$1

defines a complex nonlocality length that controls wavefront selectivity and surface scattering (Overvig et al., 2023).

In temporal and spatiotemporal photonic crystals, continuum EMT can compress Floquet sideband coupling into effective bianisotropic parameters, enabling non-Bloch band theory, generalized Brillouin zones, and criteria for the non-Hermitian skin effect (Ding et al., 2024). There the effective dispersion

$$(\hat H_\mathrm{eff}\Psi)(\mathbf r,t) = \int d^N r' \int_{0}^{t} dt'\; \mathbf h_\mathrm{eff}(\mathbf r-\mathbf r',t-t')\,\Psi(\mathbf r',t'),$$2

encodes both temporal modulation and non-Hermitian asymmetry (Ding et al., 2024).

In transmission-line metamaterials, explicit long-distance couplings yield an effective admittance

$$(\hat H_\mathrm{eff}\Psi)(\mathbf r,t) = \int d^N r' \int_{0}^{t} dt'\; \mathbf h_\mathrm{eff}(\mathbf r-\mathbf r',t-t')\,\Psi(\mathbf r',t'),$$3

and the general dispersion relation

$$(\hat H_\mathrm{eff}\Psi)(\mathbf r,t) = \int d^N r' \int_{0}^{t} dt'\; \mathbf h_\mathrm{eff}(\mathbf r-\mathbf r',t-t')\,\Psi(\mathbf r',t'),$$4

shows how nonlocal circuit parameters synthesize prescribed even dispersion curves (Ciabattoni et al., 2 Jan 2026). Time-switched activation of these nonlocal branches then enables frequency-momentum transformations at fixed $$(\hat H_\mathrm{eff}\Psi)(\mathbf r,t) = \int d^N r' \int_{0}^{t} dt'\; \mathbf h_\mathrm{eff}(\mathbf r-\mathbf r',t-t')\,\Psi(\mathbf r',t'),$$5 as an electromagnetic pulse propagates through the structure (Ciabattoni et al., 2 Jan 2026). This is a discrete realization of a temporally modulated nonlocal effective medium.

A further related line of work replaces strongly spatially nonlocal wire media by a thickness-dependent local permittivity $$(\hat H_\mathrm{eff}\Psi)(\mathbf r,t) = \int d^N r' \int_{0}^{t} dt'\; \mathbf h_\mathrm{eff}(\mathbf r-\mathbf r',t-t')\,\Psi(\mathbf r',t'),$$6, derived by averaging a nonlocal kernel over slab thickness (Yakovlev et al., 2019). Although spatial rather than temporal, this approach illustrates how a nonlocal kernel may be compressed into geometry-dependent effective parameters for practical modeling and simulation (Yakovlev et al., 2019).

These platforms collectively support applications repeatedly identified in the literature: compact devices, topologically robust transport, non-Hermitian gain/loss engineering, wavefront-selective metasurfaces, analog wave computation, and on-demand frequency-momentum transformations (Zhu et al., 9 May 2025, Overvig et al., 2023, Rizza et al., 2022, Ciabattoni et al., 2 Jan 2026).

7. Limitations, controversies, and open directions

Several limitations recur across the current formulations. First, most temporal homogenization results rely on scale separation. In acoustic temporal EMT, the derivation assumes $$(\hat H_\mathrm{eff}\Psi)(\mathbf r,t) = \int d^N r' \int_{0}^{t} dt'\; \mathbf h_\mathrm{eff}(\mathbf r-\mathbf r',t-t')\,\Psi(\mathbf r',t'),$$7, with additional conditions based on $$(\hat H_\mathrm{eff}\Psi)(\mathbf r,t) = \int d^N r' \int_{0}^{t} dt'\; \mathbf h_\mathrm{eff}(\mathbf r-\mathbf r',t-t')\,\Psi(\mathbf r',t'),$$8 for each resonance; behavior can change qualitatively when a resonance crosses the modulation scale, with the paper numerically showing a “phase transition” of effective behavior as $$(\hat H_\mathrm{eff}\Psi)(\mathbf r,t) = \int d^N r' \int_{0}^{t} dt'\; \mathbf h_\mathrm{eff}(\mathbf r-\mathbf r',t-t')\,\Psi(\mathbf r',t'),$$9 crosses $\omega$0 (Zhu et al., 9 May 2025). In electromagnetic temporal homogenization, validity is tied to $\omega$1, and agreement with full-wave Floquet theory is reported within about $\omega$2 up to $\omega$3 in representative examples (Rizza et al., 2022).

Second, higher-order or strongly noncommuting modulation effects are usually neglected. Transfer-matrix averaging in acoustic temporal EMT keeps only first-order terms in the modulation period and drops commutators such as $\omega$4 (Zhu et al., 9 May 2025). The formal two-scale electromagnetic derivation is truncated at second order in $\omega$5 (Döding et al., 27 May 2025). This suggests that strong modulation, sideband-rich dynamics, and parametric-instability regimes require beyond-EMT treatments.

Third, there is an interpretive ambiguity over what exactly should be averaged. The literature contains arithmetic averaging of $\omega$6, reciprocal averaging of $\omega$7, averaging of $\omega$8, averaging of full constitutive matrices, and effective $\omega$9-dependent corrections arising only at second order (Zhu et al., 9 May 2025, Wen et al., 2021, Rizza et al., 2022, Döding et al., 27 May 2025). This is sometimes presented as a controversy, but the more precise interpretation is that distinct microscopic embeddings of the modulation produce distinct homogenized variables. A plausible implication is that a unified classification of temporal EMT rules should be based on the operator location of the modulated parameter in the microscopic evolution equation.

Fourth, spatial and temporal nonlocality are often treated separately. Spatially homogeneous temporal media neglect spatial dispersion at the microscopic level (Rizza et al., 2022), while metasurface and transmission-line theories emphasize $\mathbf{k}$0-dependence but typically in reduced-dimensional settings (Overvig et al., 2023, Ciabattoni et al., 2 Jan 2026). A consistent theory for fully spatiotemporal metamaterials with simultaneous spatial microstructure, temporal modulation, and intrinsic resonance remains largely open. This is explicitly identified as a natural but nontrivial extension in the acoustic and electromagnetic homogenization work (Zhu et al., 9 May 2025, Döding et al., 27 May 2025).

Fifth, realizability and stability remain nontrivial. Non-Hermitian or actively modulated systems must satisfy passivity or controlled-gain constraints, and effective parameters generated by rapid switching may be difficult to implement stably (Zhu et al., 9 May 2025). In nonlocal transmission-line systems, passivity, causality, and finite propagation delays constrain which dispersion functions are physically realizable, even when the inverse-design formalism suggests wide freedom (Ciabattoni et al., 2 Jan 2026).

Current open directions are therefore well defined. They include combining temporal EMT with spatial homogenization for fully spatiotemporal crystals; extending from strictly periodic to quasi-periodic or stochastic modulation; incorporating higher-order Floquet sidebands and strong-modulation corrections; generalizing from acoustics and electromagnetics to elasticity and Willis-type systems; and developing systematic non-Bloch and topological formalisms for nonlocal, temporally modulated, non-Hermitian media (Zhu et al., 9 May 2025, Döding et al., 27 May 2025, Ding et al., 2024).

In this sense, nonlocal temporal effective medium theory is no longer limited to a temporal analogue of simple mixing formulas. It has become a broader operator-level framework for reducing time-structured, memory-bearing media to effective constitutive descriptions that remain faithful to dispersion, interfaces, finite-size effects, and, in some cases, topology and non-Hermitian wave physics.