Time-Dependent Deflection Reconstruction

• Time-dependent deflection reconstruction is a framework that recovers dynamic displacement, phase, or gradient fields from indirect tomographic or projection data using physical models and inverse problem techniques.
• The approach integrates methods such as direct geometric inversion, modal spectral expansion, harmonic CT pipelines, and quadratic estimation to address varying noise and nonlinearity in measurements.
• It underpins high-resolution metrology in diverse fields—from optics and elasticity to CMB cosmology—by employing tailored regularization and error propagation strategies for reliable reconstruction.

Time-dependent deflection reconstruction encompasses a set of methodologies and theoretical frameworks designed to recover time-varying displacement, angle, or position fields from indirect, usually tomographic or projection-based, measurements. Such reconstructions are central to disciplines ranging from optics and imaging science to continuum mechanics and gravitational wave astronomy. Across domains, the objective is to infer temporal sequences of deflection maps—either as a 3D field $d(\vec{x}, t)$ or as parameters of dynamic system evolution—by systematically integrating physical models, measurement operators, basis decompositions, and regularization strategies.

1. Mathematical Foundations and Problem Formulation

Time-dependent deflection reconstruction problems are typically cast as inverse problems where the central quantity is a time-indexed field, $h(\vec{x}, t)$, $u(\vec{x}, t)$, $z(x,y,t)$, or a phase $\phi(\vec{x}, t)$. The measured data are noisy and indirect observations (e.g., optical, diffractive, elastic, or cosmological probes) linked to the true field via a forward model—often nonlinear, nonlocal, and time-coupled. Reconstruction thus involves inversion (exact or approximate) of the combined space–time measurement operator, with explicit incorporation of physical priors, boundary and initial conditions, and measurement noise.

Notable formulations include:

• Optical grid-deflection, where slope-induced refraction yields local grid-image displacements $(\delta_x,\delta_y)$ that encode surface gradients via geometrical optics relations (Fourgeaud et al., 2018).
• Spectral time reduction in elasticity, employing a modal expansion of $\mathbf{u}(\vec{x}, t)$ in a weighted Legendre-exponential basis to project time dynamics onto coupled spatial systems (Dang et al., 15 Jun 2025).
• Analytical inversion in periodically deformed tomography by decomposing objects as $$f(\vec{x}, \varphi) = a_0(\vec{x}) + \sum_k a_k(\vec{x})\cos(k\varphi) + b_k(\vec{x})\sin(k\varphi)$$ and directly reconstructing time-harmonic coefficients from full projection datasets (Qu et al., 4 Jun 2025).
• Dynamic phase retrieval in coherent diffractive imaging, using a Markov-multiplicative update for $\Psi_{n+1}(x,y) = U_n(x,y) \Psi_n(x,y)$ and bounding $U_n$ by complex-plane sector constraints to encode temporal continuity (Tian et al., 22 May 2026).
• Time-dependent lensing estimators, in which gravitational waves impart a time-dependent deflection $h(\vec{x}, t)$0 of cosmic microwave background (CMB) features, with recovery via spatio-temporal quadratic estimators in Fourier space (Leluc et al., 14 Aug 2025).

2. Core Reconstruction Techniques

Methodologies for reconstructing time-dependent deflection fields are tightly coupled to both the physics of signal formation and the statistical structure of the measurement process.

Direct Geometric Inversion and Slope Integration

In optical deflection methods, lateral image displacements are mapped to interface slopes via simple laws such as $h(\vec{x}, t)$1. Reconstruction of $h(\vec{x}, t)$2 proceeds by spatial integration—either linewise given a reference boundary or globally via Poisson inversion:

• Line-by-line: $h(\vec{x}, t)$3
• Poisson approach: $h(\vec{x}, t)$4, typically solved with Laplacian regularization (Fourgeaud et al., 2018).

Modal and Spectral Expansion

Elastic inverse problems deploy a spectral decomposition over time: $h(\vec{x}, t)$5 where $h(\vec{x}, t)$6 (Legendre polynomial with an exponential weight), reducing the spatio-temporal inversion to a block system of spatial PDEs with coupling coefficients $h(\vec{x}, t)$7 reflecting temporal derivatives (Dang et al., 15 Jun 2025).

Harmonic and Analytical CT Pipelines

In time-resolved CT of periodic dynamics, two paradigms are prominent:

• Lock-in amplifier (LIA): Harmonic projections are separated via inner products with $h(\vec{x}, t)$8 and $h(\vec{x}, t)$9 followed by low-pass filtering, then standard filtered backprojection in each harmonic channel (Qu et al., 4 Jun 2025).
• Frequency-shifter (FS): Harmonic-weighted projections are analytically computed and summed during backprojection, efficiently pooling phase-correlated data for direct reconstruction of time-harmonic coefficients.

Feasibility Mapping by Alternating Projections

Dynamic coherent diffractive imaging constrains the inter-frame phase increment via a "circular-sector" support in the complex plane, implemented via projection operators: a measurement-domain modulus projection enforcing consistency with observed intensities, and an object-domain Euclidean projection onto the bounded sector $u(\vec{x}, t)$0. Reconstruction alternates these projections using relaxed reflect-reflect or averaged alternating reflection algorithms, iteratively refining the sequence of update maps $u(\vec{x}, t)$1 (Tian et al., 22 May 2026).

Quadratic Estimation in Spatio-temporal Frequency Space

For cosmological time-dependent deflection, measurements are decomposed into temporal and angular Fourier modes, and a quadratic estimator correlates static ($u(\vec{x}, t)$2) and dynamic ($u(\vec{x}, t)$3) components. This approach exploits the absence of primary CMB power at $u(\vec{x}, t)$4 to achieve noise reduction, estimating both gradient and curl (scalar and pseudo-scalar) components of the time-dependent deflection field (Leluc et al., 14 Aug 2025).

3. Regularization, Calibration, and Error Propagation

Reconstruction fidelity is governed by the interplay of physical model accuracy, measurement noise, and regularization strategy. Common approaches include:

• Tikhonov/Laplacian regularization applied to slope fields, as in Poisson inversion smoothing of spatial noise (Fourgeaud et al., 2018).
• Quasi-reversibility, i.e., minimization of a Tikhonov-type functional penalizing PDE residuals, data mismatch, and Sobolev norms to ensure well-posedness under noisy data (Dang et al., 15 Jun 2025).
• Low-pass harmonic filtering (in LIA pipelines) to suppress off-peak noise (Qu et al., 4 Jun 2025).
• Complex-plane support constraints, which enforce both physical passivity (amplitude ≤ 1) and temporal smoothness (bounded phase increment) in dynamic phase retrieval (Tian et al., 22 May 2026).

Calibration protocols for deflection-based surface metrology routinely involve mechanical or optical distance measurement, refractive index determination (e.g., $u(\vec{x}, t)$5 at room temperature), and in situ referencing by interferometric probes (Fourgeaud et al., 2018). Error propagation is quantified via explicit uncertainty formulas; for example, height uncertainty from optical grid deflection over a path $u(\vec{x}, t)$6 is $u(\vec{x}, t)$7, with $u(\vec{x}, t)$8m for $u(\vec{x}, t)$9 mm (Fourgeaud et al., 2018).

4. Experimental and Computational Applications

Time-dependent deflection reconstruction underpins a spectrum of experimental and computational applications:

Domain	Example System	Principal Method
Liquid films	Dewetting ridges, dynamic contact lines	Optical grid deflection + Poisson inversion (Fourgeaud et al., 2018)
Elastic solids	Anisotropic elastic wave imaging	Legendre-exponential spectral reduction and quasi-reversibility (Dang et al., 15 Jun 2025)
Medical/4D-CT	Periodically deformed biological tissue	LIA- and FS-based temporal harmonic analytic inversion (Qu et al., 4 Jun 2025)
CDI/phase imaging	Reaction-diffusion & 3D printing	Alternating projection with circular-sector prior (Tian et al., 22 May 2026)
CMB cosmology	GW-induced apparent motion	Spatio-temporal quadratic estimation (Leluc et al., 14 Aug 2025)

In wetting/dewetting, fast CMOS imaging enables sub-millisecond temporal resolution and sub-micron height quantification of moving menisci, facilitating extraction of dynamic contact angles, ridge velocities, and power-law scaling in film thickness (Fourgeaud et al., 2018). Tomographic reconstruction of periodic biological motion leverages all projections for noise reduction, matching full-dose gating performance with 25% dose via FS methods (Qu et al., 4 Jun 2025). Inelastic media, the modal time-reduction accurately recovers both geometry and amplitude of initial displacement and velocity—even under 10% boundary noise (Dang et al., 15 Jun 2025). In self-assembled photo-polymer systems, dynamic phase retrieval achieves quantitative phase recovery, revealing both reaction–diffusion patterns and growth phenomena (Tian et al., 22 May 2026). Cosmological time-dependent estimators fill the μHz GW detectability gap between PTA and space-based interferometry, although present sensitivity remains subdominant to PTA constraints (Leluc et al., 14 Aug 2025).

5. Theoretical Guarantees and Algorithmic Convergence

Mathematical analysis provides rigorous convergence and stability results in several frameworks:

• In the Legendre-exponential time-basis method for anisotropic elasticity, under standard regularity and noise conditions, the quasi-reversibility functional admits a unique minimizer, with solution error bounded by $z(x,y,t)$0 and controlled convergence to the true displacement as the noise level $z(x,y,t)$1 and regularization parameter $z(x,y,t)$2 (Dang et al., 15 Jun 2025).
• The dynamic CDI projection algorithm converges provided the per-frame phase increment remains bounded, with mean trajectory errors $z(x,y,t)$3 rad per frame and the capacity to stably integrate phase up to many $z(x,y,t)$4 wraps (Tian et al., 22 May 2026).
• In analytical harmonic CT, pooling all projections in FS-based pipelines both minimizes random noise and preserves sharp features; LIA and FS pipelines provide deterministic, parameter-free reconstruction with explicit quantitative SNR gains (LIA noise reduced by 2.5×, FS by 1.8× vs. gating) (Qu et al., 4 Jun 2025).

6. Limitations, Extensions, and Future Directions

Current methodologies assume specific model structures: local unidirectionality of slope/refraction in optical methods (Fourgeaud et al., 2018), periodicity in time-resolved CT (Qu et al., 4 Jun 2025), and bounded phase increment per frame in CDI (Tian et al., 22 May 2026). Deviations—such as non-periodic, strongly nonlinear, or spatially inhomogeneous dynamics—may require:

• Adaptive windowing, tensor decomposition, or time-resolved filtering for quasi-periodic or aperiodic cases (Qu et al., 4 Jun 2025).
• Extension of object-domain constraints or joint space–time regularization in CDI to suppress drift or cope with large increments (Tian et al., 22 May 2026).
• Nonparaxial or multi-directional inversion schemes for optical/refraction-based methods if interface slopes are not locally unidirectional (Fourgeaud et al., 2018).
• Global-trajectory regularization or Laplacian smoothing for robust error minimization in integrated surface profile maps (Fourgeaud et al., 2018, Tian et al., 22 May 2026).

In cosmological estimation, the fundamental limitation is the amplitude of the target signal and cosmic variance in the static ($z(x,y,t)$5) channel; however, the time-dependent estimator by construction excludes static variance, and thus is limited only by instrumental and foreground noise (Leluc et al., 14 Aug 2025).

7. Cross-disciplinary Impact and Generalizations

The principles of time-dependent deflection reconstruction transcend domain boundaries:

• The spatio-temporal quadratic estimator framework can be generalized from CMB lensing to any time-varying map distortion (e.g., cosmic birefringence, time-varying calibration drifts, galaxy shape distortions) (Leluc et al., 14 Aug 2025).
• Analytical inversion schemes pooling phase-correlated information over time slices are now standard in high-temporal-resolution CT, synchrotron, and in vivo imaging, with substantial impact on radiation dosage and throughput (Qu et al., 4 Jun 2025).
• Alternating projection feasibility algorithms are increasingly leveraged in dynamic phase retrieval and holography, offering a robust approach to inverse problems where physical priors can be encoded as projection sets (Tian et al., 22 May 2026).

Time-dependent deflection reconstruction thus constitutes a foundational inverse problem paradigm for high-resolution, temporally dynamic metrology and mapping across scales and physical systems.