Annealed Relaxation: Mechanisms & Applications

• Annealed relaxation is a process where controlled parameter changes systematically modify a system’s energy landscape to escape metastable states.
• It involves alternating relax steps that lower energy rapidly and replenish steps that reset accessible pathways, resulting in logarithmic annealing over time.
• The concept spans diverse domains—from atomistic mechanisms and glass dynamics to optimization and quantum protocols—enabling practical control of relaxation kinetics.

Annealed relaxation refers to a broad class of physical and algorithmic processes in which a system relaxes toward equilibrium (or a lower energy state) via an "annealing" protocol: a controlled, typically gradual modification of control parameters (temperature, strain, coupling strength, noise, regularization, etc.) that systematically modifies the system's energy landscape, barrier distribution, and accessible pathways. This concept spans materials science, statistical physics, glassy dynamics, optimization, quantum protocols, and machine learning, with a shared emphasis on unlocking, replenishing, or broadening the channels by which the system explores configuration space, thereby enabling transitions out of metastable or trapped states.

1. Atomistic Mechanisms: Replenish-and-Relax and Logarithmic Annealing

The canonical microscopic theory for annealed relaxation in disordered materials is the "^^^^1^^^^" scenario, extensively characterized for defected crystalline silicon and α-Fe via kinetic Monte Carlo and calorimetry. Two distinct types of transitions alternate:

• Relax steps: Once a configuration-space basin is accessed, a large number of low-barrier events proceed rapidly, each lowering energy by a nearly fixed amount (e.g., ≈0.6 eV in Si), analogous to collective atomic reconfigurations or defect clustering. These events correspond to experimentally observed heat release.
• Replenish steps: Upon exhaustion of low-barrier channels, the system must cross a rare, high barrier (often several times $k_BT$) to access new basins—these events do not appreciably lower energy but reset the pool of relaxation channels.

If $n(E, t)$ denotes the density of available events with barrier $E$ at time $t$, and rates are assumed Arrhenius ($k(E)=\nu \exp[-E/(k_B T)]$), the total energy relaxation follows

$E(t) \simeq E_0 - A \ln[1 + t/\tau],$

with $A$ set by the initial barrier distribution, $\mu \simeq -1.7$ for $n(E,0) \propto E^\mu$ in Si, and $\tau \sim 10^{-7}$ s. The maximal executed barrier grows log-linearly with time, $E_b^{\max}(t) \sim E_1 + \beta \ln(t/\tau_1)$, and the relaxation rate is progressively limited by the need to replenish, leading to logarithmic annealing over many decades in $t$ (Béland et al., 2013, Béland et al., 2014).

2. Glasses and Amorphous Materials: Dynamic Pathways and Kinetic Implications

In amorphous solids and glasses, annealed relaxation phenomena are central to structural evolution, aging, and the glass transition. Kinetically constrained models (e.g., East or facilitated spin models) for ultrastable glasses reveal two concurrent mechanisms:

• Ballistic front propagation: A highly mobile surface or interface, excited or annealed, seeds a front that propagates rapidly into the bulk; the front speed $v(T)$ exhibits super-Arrhenius slowing with decreasing $T$.
• Bulk nucleation-and-growth: In deep, ultrastable bulk, relaxation requires rare, spontaneous nucleation of excitations, growing via facilitated dynamics. The Kolmogorov–Johnson–Mehl–Avrami law applies, with the transformed fraction $f_{\rm bulk}(t) = 1 - \exp[-k t^n]$ and a bulk relaxation time exponentially large in the activation barrier.

A crossover length $L_c \sim v(T) \tau_{\rm bulk}(T)$ distinguishes thickness regimes dominated by either surface- or bulk-driven annealed relaxation (Gutierrez et al., 2016). These models unify the observed kinetic length scale separation in ultrastable glass melting.

3. Mechanical Annealing and Cyclic Loading Protocols

Annealed relaxation under mechanical driving occurs in glasses, jammed solids, and metallic glasses subjected to cyclic shear or stress. Typical protocols include:

• Constant-amplitude cyclic shear: Repeated deformation at subyield amplitude leads to an exponential or logarithmic relaxation of potential energy, quantified as $U(n) = U_\infty + A \ln n$ (metallic glasses (Priezjev, 2024)), or $$U_{\min}(n) = U_\infty + [U_0 - U_\infty] e^{-n/\tau_E}$$ (binary Yukawa/LJ glasses (Priezjev, 2017)). The characteristic time or amplitude dependence governs the pace of approach to more stable, lower-energy configurations.
• Ring-down annealing protocols: Stepwise reduction of amplitude ("mechanical cooling") can erase memory content and reach a more isotropic, minimally anisotropic state (Keim et al., 2021).

Structural relaxation involves progressive reduction of collective, irreversible atomic rearrangements, often measured by nonaffine displacement metrics $D^2$. The ultimate steady state achievable before yielding is determined by a critical strain amplitude $\gamma_c$, above which reversible annealing ceases and shear banding or flow occurs (Priezjev, 2024, Priezjev, 2017). Hybrid Monte Carlo methods embedding cyclic shear as trial moves enable true equilibrium sampling and suppress long-time crystallization (Das et al., 2018).

4. Annealed Relaxation in Algorithmic and Quantum Contexts

Annealed relaxation principles are increasingly harnessed in optimization, machine learning, and quantum algorithms:

• Optimal Transport: Annealed Sinkhorn algorithms increase the entropic-regularization parameter $\beta_t$ in a nondecreasing schedule, solving a sequence of progressively less relaxed optimal transport problems. The relaxation error scales as $O(\beta_t - \beta_{t-1})$, with best error achieved for $\beta_t \sim \sqrt{t}$. Debiased variants reduce this error to sublinear rates (Chizat, 2024).
• Relaxation-based importance sampling: For estimating rare event probabilities, annealed schedules (either through an increasing exponent $\beta_t$ or decreasing covariance $\lambda_t$) interpolate from an easy, highly relaxed problem to the true (hard) target, with intermediate resampling and MCMC enhancing efficiency. These methods generalize and unify sequential importance sampling, subset simulation, and annealed importance sampling methods (Xian et al., 2023).
• Quantum cooling protocols: In quantum simulation, annealed relaxation refers to driving the system–bath coupling through a controlled, decreasing schedule—strong at early cycles for rapid entropy extraction, weak at late stages to preserve ground-state fidelity. Time-modulated baths further accelerate relaxation toward the ground state, with empirical performance verified in transverse-field Ising models (Xu, 9 Jan 2025).
• Reverse annealing with continuous relaxation: Relaxation-assisted reverse annealing uses continuous solutions from linear or quadratic programming relaxations as initializations for quantum or classical annealing, improving convergence and locality in combinatorial optimization tasks (Haba et al., 3 Jan 2025).

5. Experimental Probes and Structural Evolution in Disordered Solids

Experimental studies of annealed relaxation employ diverse probes:

• Nanocalorimetry: Measures heat release during relax steps, revealing the density and distribution of accessible barriers in materials like irradiated Si (Béland et al., 2013).
• Mechanical spectroscopy and calorimetry: Suppression of secondary ($\beta$-) relaxations via sub-$T_g$ annealing is reflected as the diminishment of the excess wing in loss modulus spectra and modification of crystallization kinetics (GeTe (Makareviciute et al., 15 Oct 2025), densified SiO$_2$ (Cornet et al., 2024)).
• Raman spectroscopy: Structural relaxation in amorphous semiconductors is traced via bond-angle disorder; relaxation and derelaxation pathways (e.g., Si vs. Si:H) are observed as shifts in the Raman-active TO band and quantified heat flows (Kail et al., 2010).
• Scanning probes and magnetic force microscopy: Hardware realization of annealed relaxation in nanomagnet arrays demonstrates the emulation of simulated annealing via time-varying global strain, enabling the system to escape metastable traps (Abeed et al., 2019).

Mechanistically, many systems display a two-step relaxation via transitory network inhomogeneity (e.g., short-lived increase in three-membered rings in silica during density recovery (Cornet et al., 2024)), with relaxation kinetics well described by stretched exponentials (KWW), $\tau = \tau_0 \exp(E_a/RT)$.

6. Unified Theoretical Frameworks and State-Space Dynamics

Modern state-space and random map models formalize annealed relaxation as iterated maps on a high-dimensional (potentially discrete) space of locally stable configurations:

• Random and constrained maps: The convergence properties and memory content of iterated forward/reverse maps (U, D) reveal how physical constraints (locality, threshold structure, return-point memory) control the number of accessible cycles, tail lengths, and component sizes in system response (Mungan et al., 2019, Keim et al., 2021).
• Annealed trees and polymer melts: In the annealed-tree model for polymer ring melts, the relaxation time, viscosity, and diffusion constants scale with mass $N$ due to the backbone's thermal annealing freedom, explaining the unique dynamics of ring polymers (Smrek et al., 2014).

These perspectives support a unified view in which annealed relaxation arises from the interplay of an evolving barrier landscape, stochastic or deterministic protocol schedules, and intricate constraints (kinetic, topological, thermodynamic) governing the accessibility of relaxation pathways. They enable systematic acceleration, control, and understanding of relaxation phenomena in both physical and algorithmic systems.