Linear Memory Effect: Theory & Applications

Updated 9 January 2026

Linear memory effect is defined as the explicit dependence of a system’s current state on its entire past through a linear kernel.
It emerges in physical observables such as shear stress and gravitational wave strains, and is analyzed using integro-differential equations and Markovian mappings.
Experimental and computational studies, including colloidal glass rheometry and gravitational wave detection, validate its role in nonequilibrium dynamics.

The linear memory effect encompasses a class of physical phenomena and mathematical structures in which a system’s current state exhibits explicit dependence on its entire past via a linear kernel. This effect is fundamental in nonequilibrium statistical mechanics, condensed matter, gravitational wave physics, and dynamical systems. The linear memory effect appears in physical observables ranging from the shear stress in colloidal glasses subject to two-step mechanical protocols, to the permanent metric offsets induced by gravitational radiation in scattering events, and is rigorously treated via integro-differential equations in both continuous and discrete time. Its characterization relies on the system’s linear response function (“memory kernel”), analytic structure in frequency space (such as zero-frequency poles), and mappings to Markovian dynamics in higher dimensions. The following sections survey the principal definitions, mathematical frameworks, experimental and computational methodologies, dimensionality effects, and selected applications of linear memory phenomena.

1. Mathematical Structure and Definitions

Linear memory effects are rigorously described by evolution equations incorporating convolution terms, representing hereditary dependence on past states. In the continuous-time autonomous case, the generic linear memory equation reads

$\dot x(t) = \int_0^t K(t-\tau) x(\tau) \, d\tau + f(t),$

with $K$ the memory kernel (often exponentially decaying, see (Brandner, 2024)), and $f(t)$ an optional source. For observable quantities in nonequilibrium systems, such as stress or temperature, the relaxation follows

$\sigma(t) = \int_{0}^{t} \chi(t-t') \dot{\gamma}(t') d t',$

where $\chi(t)$ is the stress-memory kernel as in Kovacs–like effects (Mandal et al., 4 Jan 2025).

In discrete time, the analogous recurrence is

$X_{n+1} = V X_n + \sum_{m=1}^{n} K_m X_{n-m},$

with $V$ instantaneously propagating the state and $K_m$ encoding $m$ -step memory (Meyer et al., 30 Oct 2025). In both formulations, linearity in the variable of interest is essential; memory effects due to quadratic or higher order terms are beyond the strictly linear regime.

2. Physical Realizations: Systems and Observable Manifestations

Table: Physical Systems Exhibiting Linear Memory Effects

System/Process	Observable	Kernel/Signature
Sheared colloidal glass (Mandal et al., 4 Jan 2025)	Shear stress $\sigma(t)$	Double-log law, non-monotonic hump
Athermal lattice model (Plata et al., 2017)	Granular temperature $T(t)$	Markov master equation, Kovacs hump
Gravitational wave sources (Favata, 2010, Hait et al., 2022)	GW strain $h_{ij}^{TT}(t)$	Permanent DC offset, $1/\omega$ pole
Supernova neutrino shell (Kodwani et al., 2016)	Relative separation $l(t)$	Longitudinal step, time linear transverse
Markov reduced process (Stephan et al., 2018)	Projected variable $u(t)$	Exponential kernel, equilibrium via projection

In mechanically-driven amorphous solids, the Kovacs-like memory effect manifests as a non-monotonic evolution of macroscopic stress under a two-step shear protocol, quantifiable within the linear response regime by the convolution of the strain-rate with a logarithmically decaying response kernel (Mandal et al., 4 Jan 2025). In stochastic athermal lattices, higher-order cumulants (e.g., excess kurtosis) play a role in storing memory, and the Kovacs hump is universally predicted by the moment equations derived from the system's master equation (Plata et al., 2017).

Gravitational wave memory provides a paradigmatic example of linear memory at relativistic scales: the passage of a gravitational wave pulse sourced by mass ejection or scatter imparts an unrecoverable displacement to test masses, mathematically encoded by zero-frequency poles in the strain’s Fourier domain (Favata, 2010, Hait et al., 2022). The effect is physically manifest in non-periodic, unbound motions, hyperbolic encounters, and supernova neutrino shells, producing a permanent separation change in interferometer arms or pulsar timing residuals (Kodwani et al., 2016).

3. Theoretical Frameworks and Markovian Embedding

A key insight is the equivalence between certain classes of linear memory equations and projections of finite-dimensional Markov processes. This connection, formalized in (Stephan et al., 2018), asserts that any memory kernel constructed from nonnegative exponential combinations (loops) can be realized as the “visible” component of a higher-dimensional Markov chain: $\dot u(t) = -a u(t) + \int_{0}^{t} K(t-s) u(s) ds,$ maps to the first component of $\dot p(t) = Q p(t)$ , with $p(t) \in \mathbb{R}^{N+1}$ and explicit generator $Q$ . This interpretation yields explicit computation of equilibrium states, long-time asymptotics, and spectral gaps, allowing application of Markov semigroup theory to initial non-Markovian systems (Stephan et al., 2018).

In weak-memory regimes—where the memory kernel decays faster than the typical system rate $v<k$ with $M$ the kernel supremum—a perturbative expansion delivers a local approximation to the evolution generator, with error bounds and convergent series for effective dynamics (Brandner, 2024, Meyer et al., 30 Oct 2025). The generator $L$ satisfies uniform bounds on the deviation from the memoryful dynamics, enabling controlled simplification of otherwise history-dependent evolution.

4. Frequency-Domain Analysis and Gravitational Memory

Linear memory in gravitational wave phenomena is tightly linked to the analytic properties of waveform templates in the frequency domain. For hyperbolic encounters, the strain’s Fourier transform $\tilde h(\omega)$ exhibits a $1/\omega$ pole at $\omega \to 0$ , producing a step function (“DC offset”) in the time domain—a signature of permanent displacement (Hait et al., 2022, Dutta et al., 3 Nov 2025). The detailed field-theoretic analysis reveals additional non-analytic corrections, notably $\ln \omega$ terms arising from Hankel function expansions, indicating subtle tails in the post-scattering waveform (Hait et al., 2022).

Crucially, for exactly parabolic orbits ( $e=1$ ), such a $1/\omega$ pole is absent; instead, fractional singularities ( $\omega^{-1/3}$ ) dominate, and no net memory step appears. This demonstrates the breakdown of naive limiting procedures from hyperbolic or elliptic cases and clarifies the need for careful parametrization near the marginally bound regime (Dutta et al., 3 Nov 2025).

Odd spacetime dimensions behave differently: Satishchandran & Wald (Satishchandran et al., 2017) prove the absence of gravitational linear memory for all odd $d \ge 5$ , while scalar and electromagnetic fields show peculiar tail behaviors, e.g., unbounded displacement memory but vanishing momentum memory for $d=5$ .

5. Experimental and Computational Protocols

In experimental soft matter (e.g., PNIPAM microgel colloidal glasses), the linear memory effect was characterized by stress-controlled rheometry with high temporal resolution, mapping macroscopic stress relaxation to a fitted double-log kernel (Mandal et al., 4 Jan 2025). Simulations of athermal bidisperse disks corroborate these results, with linearity observed up to critical strain amplitudes and breakdowns identified via direct imaging of non-affine rearrangements.

Numerical integration of moment equations and Monte Carlo simulation of discrete master equations validate theoretical predictions of Kovacs-like humps and detailed memory kernel forms in athermal lattice models (Plata et al., 2017).

In gravitational wave detection, direct observation of linear memory (as a DC offset) remains challenging for ground-based interferometers due to frequency sensitivity limitations, but space-based detectors (LISA) and pulsar-timing arrays are potentially sensitive to the permanent memory imprinted by large-scale astrophysical sources (Favata, 2010, Kodwani et al., 2016).

6. Robustness, Limitations, and Dimensionality Effects

Strict linearity is observed within narrow parameter regimes: in colloidal glasses, linear response theory is accurate up to strains of a few percent, beyond which non-affine flows and plasticity distort the predicted memory kernels (Mandal et al., 4 Jan 2025). In discrete and continuous-time weak-memory scenarios, the reduction to effective first-order dynamics holds rigorously only when coupling constants and spectral decay rates satisfy precise inequalities (Brandner, 2024, Meyer et al., 30 Oct 2025).

Dimensional constraints are fundamental in field theory: gravitational linear memory vanishes in odd $d \geq 5$ , scalar and electromagnetic cases display unbounded or absent tails depending on $d$ , and all effects are absent for slowly moving, finite-duration sources in higher dimensions (Satishchandran et al., 2017). These mathematical results delimit the physical observability and conceptual reach of the linear memory effect.

7. Selected Applications and Extensions

Linear memory effects underpin data interpretation in ferroelectric field-effect transistors (FeFETs), where the memory window (threshold voltage separation) is linearly proportional to remanent polarization in the weak-polarization regime, with explicit crossover at larger $P_r$ to saturation via coercive field (Toprasertpong et al., 2022).

The theoretical framework admits generalization to stochastic Floquet dynamics, quantum collisional models, and coarse-graining procedures for Markov chains, with rigorous fixed-point equations for the effective propagator and slippage matrices (Meyer et al., 30 Oct 2025). Systematic perturbation schemes and derivative-moment expansions furnish practical tools for reducing non-Markovian memory equations to tractable Markovian models (Brandner, 2024, Stephan et al., 2018).

A plausible implication is the broad relevance of linear memory phenomena for emergent mesoscale dynamics, signal processing in astrophysical detectors, and the design of memoryful devices in quantum technologies.

References: (Mandal et al., 4 Jan 2025, Plata et al., 2017, Favata, 2010, Dutta et al., 3 Nov 2025, Hait et al., 2022, Brandner, 2024, Meyer et al., 30 Oct 2025, Stephan et al., 2018, Kodwani et al., 2016, Satishchandran et al., 2017, Toprasertpong et al., 2022)