Dynamic Optical Perturbations

• Dynamic optical perturbations are time-dependent changes in optical observables (phase, amplitude, path difference, etc.) that influence both measurement signals and system performance.
• They enable advanced imaging and sensing techniques by introducing controlled modulation or compensating for distortions, as demonstrated in methods like rolling-phase full-field OCT.
• Engineered perturbations in metasurfaces and adaptive optics facilitate real-time beam steering and optical compensation with response times under 5 ms, enhancing system functionality.

Searching arXiv for papers on dynamic optical perturbations and related optical modulation/imaging/control topics. Fetching arXiv metadata to ground the article in current literature. $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$58 Dynamic optical perturbations are time-dependent changes in optical phase, amplitude, optical path difference, refractive index, scattering response, or modal content, whether introduced deliberately as a control variable or arising from environmental and material dynamics. In contemporary optics, the term spans coherence-gated microscopy with externally imposed phase ramps, intra-scan interferometric baseline oscillations, time-varying transmission matrices in scattering media, dynamic turbulence in free-space propagation, nanometer-scale cavity-gap modulation in metasurfaces, hydrogen-loading-induced permittivity changes, and particle-induced perturbations in non-Hermitian resonators. The same perturbation can therefore be treated as a measurable signal, a nuisance to be rejected, or an engineered degree of freedom for functionality (Monfort et al., 14 Jan 2025, Zhou et al., 2024, Meng et al., 2021).

1. Conceptual scope and governing descriptions

A unifying feature of dynamic optical perturbations is that they are modeled as explicitly time-varying modifications of an optical observable or of the operator that maps input fields to output fields. In dynamic full-field OCT, the perturbation enters as a reference-arm phase ramp $\phi_{\rm ref}(t)$ in the interferometric intensity

$$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$

so that sample fluctuations $\phi_s(t)$ and externally applied modulation coexist in the same measurement model (Monfort et al., 14 Jan 2025). In wavelength-scanning interferometry, the perturbation is a time-varying optical path difference $L(t)=L_0+\delta L(t)$ encoded during a single spectral sweep and recovered by demodulating the recorded spectrum rather than assigning one OPD value to the entire scan (Ushakov et al., 2019).

In dynamic scattering and wavefront-control problems, the perturbation is often expressed as a time-dependent transmission operator. One representative model writes

$$E_{\rm out}(m,t)=\sum_{n=1}^N T_{mn}(t)\,E_{\rm in}(n),$$

with $T(t)$ changing as the medium decorrelates; inserting a fixed modulation matrix $M$ produces an effective operator $A(t)=M\,T(t)$ (Li et al., 21 Aug 2025). In adaptive-optics and beam-control settings, perturbations are recast in dynamical-systems form, for example as AR2 models for turbulence and structural vibrations, or as state-space disturbances in Kalman filtering and subspace identification (Jaufmann et al., 16 Sep 2025, Haber et al., 2023).

A second organizing distinction is between perturbation-as-signal and perturbation-as-distortion. In metrology and microscopy, the objective may be to estimate picometer-scale OPD excursions or nanoscale intracellular dynamics from the perturbation itself (Ushakov et al., 2019, Monfort et al., 14 Jan 2025). In scattering, turbulence, and optical communications, the same class of time dependence is typically an impairment to be compensated (Li et al., 2020, Zhou et al., 2024). In tunable photonic structures and materials, perturbations are intentionally driven to obtain programmable optical transfer functions (Meng et al., 2021, Palm et al., 2018).

2. Interferometric perturbations in microscopy and precision metrology

A particularly explicit implementation of dynamic optical perturbation appears in Rolling-Phase dynamic full-field OCT. The method uses a Linnik interferometer with LED source at $\lambda_0=810$ nm and $\Delta\lambda=25$ nm, identical oil-immersion objectives $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$0, and a reference mirror on a piezoelectric transducer. The acquisition consists of $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$1 frames at $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$2 Hz, giving $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$3 s, while the reference phase is ramped linearly from $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$4 to either $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$5 or $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$6 according to $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$7 (Monfort et al., 14 Jan 2025). The short-time difference metric

$$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$8

is then averaged to form a dynamic-brightness channel, while the static reflectivity is recovered from the FFT amplitude at the carrier frequency $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$9. Because $\phi_s(t)$0 sweeps by at least $\phi_s(t)$1, the unknown static phase map $\phi_s(t)$2 is averaged out, which removes the strong spatial nonuniformity that characterizes standard DFFOCT. Experimentally, the method removes fringe artifacts, reduces speckle contrast, drives residual $\phi_s(t)$3-dependent noise below $\phi_s(t)$4 when the sweep is at least $\phi_s(t)$5, and achieves comparable dynamic-contrast SNR with approximately $\phi_s(t)$6 fewer frames. On monkey retinal explants it resolves photoreceptor outer segments, clear nuclear boundaries, and intranuclear activity, while extracting static and dynamic contrast from the same dataset (Monfort et al., 14 Jan 2025).

A closely related but metrologically distinct use of dynamic perturbations appears in wavelength-scanning interferometry for baseline measurement. There, a low-finesse Fabry–Perot spectrum

$\phi_s(t)$7

is recorded during a wavelength sweep, and intra-scan OPD variations are retrieved by Hilbert-transform demodulation on a uniform frequency grid (Ushakov et al., 2019). The procedure first estimates a mean OPD $\phi_s(t)$8 by least-squares fitting, then unwraps the analytic-signal phase, subtracts the non-perturbed term, and converts the residual phase to instantaneous $\phi_s(t)$9. An analytical model gives detectability bounds in perturbation amplitude and frequency, including the lower-frequency requirement $L(t)=L_0+\delta L(t)$0 and a Carson-rule-style constraint coupling $L(t)=L_0+\delta L(t)$1, $L(t)=L_0+\delta L(t)$2, and carrier $L(t)=L_0+\delta L(t)$3. In experiment, a swept-laser spectrometer with wavelength range $L(t)=L_0+\delta L(t)$4–$L(t)=L_0+\delta L(t)$5 nm, $L(t)=L_0+\delta L(t)$6 pm, $L(t)=L_0+\delta L(t)$7 points, and $L(t)=L_0+\delta L(t)$8 s successfully tracked oscillations up to $L(t)=L_0+\delta L(t)$9 Hz with amplitudes as small as $$E_{\rm out}(m,t)=\sum_{n=1}^N T_{mn}(t)\,E_{\rm in}(n),$$0 nm in a single sweep. The measured noise spectral density was $$E_{\rm out}(m,t)=\sum_{n=1}^N T_{mn}(t)\,E_{\rm in}(n),$$1 pm/$$E_{\rm out}(m,t)=\sum_{n=1}^N T_{mn}(t)\,E_{\rm in}(n),$$2, versus a theoretical $$E_{\rm out}(m,t)=\sum_{n=1}^N T_{mn}(t)\,E_{\rm in}(n),$$3 pm/$$E_{\rm out}(m,t)=\sum_{n=1}^N T_{mn}(t)\,E_{\rm in}(n),$$4, and the integrated RMS noise over $$E_{\rm out}(m,t)=\sum_{n=1}^N T_{mn}(t)\,E_{\rm in}(n),$$5 was approximately $$E_{\rm out}(m,t)=\sum_{n=1}^N T_{mn}(t)\,E_{\rm in}(n),$$6 pm (Ushakov et al., 2019).

These two examples illustrate a recurrent principle: dynamic perturbation can be introduced slowly enough to linearize or decorrelate an otherwise phase-sensitive measurement, yet still retain sufficient temporal structure to recover the underlying signal of interest.

3. Scattering, turbulence, and one-to-many compensation

In dynamically perturbed scattering media, the central difficulty is that compensation methods based on transmission matrices, iterative wavefront shaping, or optical phase conjugation ordinarily assume quasi-static conditions. One line of work addresses this by deriving an explicit metric linking optical performance to perturbation magnitude. For binary-amplitude wavefront shaping through strong scattering, the “square-rule” states that if a fraction $$E_{\rm out}(m,t)=\sum_{n=1}^N T_{mn}(t)\,E_{\rm in}(n),$$7 of SLM pixels are incorrect, then

$$E_{\rm out}(m,t)=\sum_{n=1}^N T_{mn}(t)\,E_{\rm in}(n),$$8

equivalently

$$E_{\rm out}(m,t)=\sum_{n=1}^N T_{mn}(t)\,E_{\rm in}(n),$$9

This gives a real-time estimate of how far the current mask has drifted from the optimum, and it underlies the Dynamic Mutation Algorithm, which adaptively mutates the mask according to the inferred error rate (Li et al., 2020). Under simulated and experimental sudden perturbations, including multimode-fiber rotation, the method recovered the original PBR without stopping the run: a $T(t)$0 fiber rotation caused a drop of about $T(t)$1, followed by recovery in approximately $T(t)$2 measurements, whereas GA remained trapped and CSA recovered more slowly (Li et al., 2020).

A different strategy replaces per-realization adaptation with fixed optical modulation learned across a family of perturbations. In real-time imaging through dynamic scattering media, a static modulation operator $T(t)$3 is optimized so that $T(t)$4 approximately inverts many scattering realizations simultaneously. The physical feasibility is tied to the optical shower-curtain effect, and the reported operational regime is typically within $T(t)$5–$T(t)$6 transport mean free paths (Li et al., 21 Aug 2025). Using optical diffraction neural networks trained on simulated media, the method reconstructs images in real time with light-speed optical inference, achieves $T(t)$7 Hz imaging of moving objects in dynamic scattering media with decorrelation times $T(t)$8 ms, and uses speckle decorrelation as a source of temporal averaging rather than treating it only as a failure mode. The same work explicitly notes that the fixed-mask approach fails for thick volumetric scattering once the shower-curtain correlation is lost (Li et al., 21 Aug 2025).

Dynamic atmospheric turbulence poses an analogous problem in coherent free-space links. An automatic mitigation scheme based on four-wave mixing in a GaAs photorefractive crystal records a turbulence-distorted probe, regenerates a Gaussian reference from that probe, and combines both with a high-speed Gaussian data beam to create a pre-distorted phase-conjugate signal (Zhou et al., 2024). In the modal formalism, reciprocity and losslessness imply $T(t)$9, so the return propagation cancels the forward distortion when the OPC response is sufficiently fast. Experimentally, the system operates with an $M$0-Gbit/s QPSK data beam and response time $M$1 ms, while turbulence is characterized by Greenwood frequency $M$2. At $M$3 Hz, the method yields about $M$4 dB reduction in mean SMF-coupling loss; across the tested range up to about $M$5 Hz, the mitigated beam retains predominantly $M$6 content and all $M$7 realizations remain below the FEC threshold, unlike the unmitigated case (Zhou et al., 2024).

Passive compensation has also been proposed. In turbulence-impacted PMMA rods modeled as coupled anharmonic Lorentz dipole oscillators, the relevant perturbation quantity is

$M$8

which measures the inertial resistance of synchronized dipole motion to rapid field redistribution (Sadhukhan et al., 5 Nov 2025). In a pseudo-random phase-plate experiment, the mean scintillation index changed from $M$9 for turbulence only to $A(t)=M\,T(t)$0 with one rod and $A(t)=M\,T(t)$1 with two rods, corresponding to approximately $A(t)=M\,T(t)$2 and $A(t)=M\,T(t)$3 reductions (Sadhukhan et al., 5 Nov 2025).

A common misconception is that compensation in dynamic scattering must remain one-to-one and re-optimized for each realization. The literature now includes both adaptive and one-to-many strategies, but their validity is sharply regime-dependent.

4. Engineered perturbations in metasurfaces, antennas, and tunable materials

Dynamic optical perturbations are also used as an intentional actuation mechanism in reconfigurable photonic hardware. In a MEMS-based optical metasurface platform, an electrically tunable air gap between a nanobrick metasurface and a gold mirror acts as the perturbation coordinate. The stack is modeled as a Fabry–Pérot resonator with complex reflection coefficient

$A(t)=M\,T(t)$4

from which the reflected amplitude $A(t)=M\,T(t)$5 and phase $A(t)=M\,T(t)$6 follow (Meng et al., 2021). The OFF-state gap is about $A(t)=M\,T(t)$7 nm and the ON-state gap approaches $A(t)=M\,T(t)$8 nm in simulation and about $A(t)=M\,T(t)$9 nm experimentally. For a representative nanobrick at $\lambda_0=810$0 nm, the phase varies from approximately $\lambda_0=810$1 at $\lambda_0=810$2 nm to $\lambda_0=810$3 at $\lambda_0=810$4 nm and $\lambda_0=810$5 at $\lambda_0=810$6 nm; experimentally, the platform provides polarization-independent beam steering and two-dimensional focusing with measured modulation efficiencies around $\lambda_0=810$7, relative optical bandwidth around $\lambda_0=810$8 near $\lambda_0=810$9 nm, and response times $\Delta\lambda=25$0 ms and $\Delta\lambda=25$1 ms (Meng et al., 2021).

In silicon nitride optical leaky-wave antennas, the perturbation is periodic semiconductor loading of the waveguide. The demonstrated device uses $\Delta\lambda=25$2 crystalline Si perturbations with period $\Delta\lambda=25$3m, each $\Delta\lambda=25$4 nm long and $\Delta\lambda=25$5 nm high, producing a directive radiation pattern at $\Delta\lambda=25$6 nm with measured maximum intensity at $\Delta\lambda=25$7 relative to the waveguide axis and half-power beam width around $\Delta\lambda=25$8 (Zhao et al., 2015). Dynamic control arises from electrically induced index changes in the perturbations. Theoretical analysis reported $\Delta\lambda=25$9 with depletion-mode bandwidth $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$00 GHz, $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$01 with injection-mode bandwidth about $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$02 MHz, and Ge-based Franz–Keldysh modulation with $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$03 at $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$04 kV/cm near $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$05 nm (Zhao et al., 2015).

Hydrogenation provides a chemically driven perturbation of optical constants. A systematic in situ study of Pd, Mg, Zr, Ti, and V across $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$06–$$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$07 nm used an environmental chamber that combined quartz-crystal-microbalance mass measurements with ellipsometry up to $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$08 bar of hydrogen gas (Palm et al., 2018). The partially hydrided state was modeled with a Bruggeman effective-medium approximation, while device demonstrations showed dynamic behavior of directly photonic relevance: five orders of magnitude change in reflectivity, resonance shifts of more than $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$09 nm, and relative transmission switching of more than $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$10 (Palm et al., 2018). These results place dynamic perturbations on the same footing as static material dispersion engineering: the optical transfer function itself becomes a trajectory in material-state space.

5. Estimation, forecasting, and control of perturbation dynamics

Once perturbations become fast relative to acquisition or control latency, estimation theory becomes central. A data-driven framework for optical-spot dynamics models beam motion with

$$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$11

and identifies both disturbance statistics and dynamical structure from measured spot trajectories (Haber et al., 2023). The approach combines autocovariance least squares for estimating $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$12 and $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$13, Kalman-filter tuning, subspace identification, and spectral factorization for laboratory disturbance synthesis. On an experimental bench with a piezo tip-tilt mirror, piezo linear actuator, and CMOS camera, a subspace-identified model achieved $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$14 VAF in the tip-tilt mirror test, with residual autocorrelation showing only $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$15 outliers; for the linear stage, the reported VAF was about $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$16 (Haber et al., 2023).

For large-telescope adaptive optics, the disturbances of interest are atmospheric turbulence and structural vibrations. One simulation study used AR2 models for low-order Zernike modes and tip–tilt vibration modes, stacked together with deformable-mirror delays in a six-dimensional per-mode state (Jaufmann et al., 16 Sep 2025). A multi-rate Kalman filter fuses LGS measurements at $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$17 kHz with NGS measurements at $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$18 Hz every fifth sample, while a separate Gaussian-process-regression observer estimates structural perturbations from wavefront data alone. In the reported performance results, pure LGS-only tip–tilt residuals of RMS approximately $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$19 nm were reduced to about $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$20 nm after Kalman filtering, and the multi-rate LGS+NGS filter produced an additional reduction of about $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$21 nm in low-frequency residuals. The offline GPR observer reduced vibration-estimation RMS from about $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$22m in closed loop to about $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$23m (Jaufmann et al., 16 Sep 2025).

Forecasting methods that do not assume a mechanistic model have also been applied to aero-optic wavefronts. Optimized dynamic mode decomposition fits a globally consistent exponential model

$$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$24

and on AAOL-T wavefront data produced eigenvalues lying almost entirely on the imaginary axis, so that modal evolution is oscillatory rather than spuriously decaying (Sahba et al., 2021). By contrast, exact DMD yielded many small negative real parts corresponding to modal half-lives typically in the $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$25–$$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$26s range, with mean about $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$27s and amplitude-weighted mean about $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$28s. The significance is operational rather than merely descriptive: future-state prediction remains coherent over the AO latency horizon when the learned spectral representation respects the perturbation’s oscillatory structure (Sahba et al., 2021).

These control-oriented formulations treat perturbation not as an isolated optical defect but as a stochastic or deterministic dynamical process coupled to sensing, actuation, and computational latency.

6. Emergent phenomena, adversarial uses, and limits of the concept

Dynamic optical perturbations are not confined to imaging or control; they also produce new physical effects and new attack surfaces. In optical tweezers integrated with differential dynamic microscopy, the perturbation is the controlled motion of the trap center $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$29 over a total stroke $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$30m with pause $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$31 s, imposing strain rates from $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$32 to $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$33 (Peddireddy et al., 2022). The resulting polymer response is mapped through the image-structure function

$$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$34

and ring-linear DNA blends show maximal affine alignment, superdiffusive scaling, and elastic memory at $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$35, between predicted entanglement rates $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$36 and $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$37. Adding microtubules suppresses this resonance while increasing elastic retention (Peddireddy et al., 2022). Here, optical perturbation is a means of probing nonlinear stress propagation rather than of forming an image.

In computer vision, dynamic optical perturbation has been weaponized. The EvilEye system mounts a transparent display and auxiliary sensor directly in front of a camera and models the camera–display transfer function as

$$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$38

approximated by a learned UNet-based “TNet” (Han et al., 2023). In physical traffic-sign experiments, the attack maintained classifier ASR of about $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$39 up to $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$40 lux indoors and about $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$41 at $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$42 lux; for detectors it remained around $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$43 up to $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$44 lux outdoors and about $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$45 at $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$46 lux, while a neutral-density filter preserved more than $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$47 ASR under direct sun at $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$48–$$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$49 lux. Dynamic class-specific perturbations achieved average ASR around $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$50–$$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$51 across sampled routes, compared with about $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$52–$$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$53 for a single static perturbation (Han et al., 2023). This establishes that optical perturbation can target the sensor pipeline itself rather than the propagating field.

At a more fundamental level, perturbations can alter the symmetry class of an optical system. In a spinning two-mode resonator, nanoparticle scattering, gain/loss, and Sagnac detuning define a non-Hermitian Hamiltonian whose eigenvalues

$$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$54

can be tuned from anti-$$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$55 behavior to a quasi-closed Hermitian regime by adjusting particle-induced complex couplings and rotation (Zhang et al., 2024). In another setting, circularly polarized scattering from a sphere near an interface converts spin angular momentum into orbital circulation, producing a lateral force whose sign depends on handedness and whose magnitude is about $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$56 of the usual radiation-pressure force at small separations (Sukhov et al., 2015). These cases show that a perturbation need not be merely corrective or disruptive; it can change the effective topology or momentum balance of the optical system.

A recurring limit across the literature is timescale matching. Rolling-phase OCT requires a highly linear piezo scan over at least $$I(t)=I_r + I_s + I_{\rm inc} + 2\sqrt{I_r\,I_s}\cos\!\bigl[\phi_s(t)+\phi_0+\phi_{\rm ref}(t)\bigr],$$57 and a sweep rate slower than the fastest sample fluctuations of interest yet faster than slow drifts (Monfort et al., 14 Jan 2025). Fixed-modulation imaging through scattering is limited to thin media with surviving shower-curtain correlations (Li et al., 21 Aug 2025). OPC bandwidth is bounded by photorefractive response time (Zhou et al., 2024). Wavefront-shaping compensation fails when medium decorrelation outruns SLM and detector bandwidth (Li et al., 2020). This suggests that “dynamic optical perturbation” is less a single technique than a regime concept: it names the interval in which optical dynamics are too fast to ignore, yet still slow enough to sense, model, or exploit.