Generalized Rate-Dependent Hysteresis Models

• Generalized rate-dependent hysteresis models are frameworks that capture nonlinear memory effects by incorporating time or fractional derivatives into classical constitutive relations.
• They integrate generalized play operators, fractional-order formulations, and data-driven neural network methods to model and predict dynamic phenomena such as loop scaling and strain stiffening.
• Calibration techniques like multi-branch fitting and evolutionary algorithms allow these models to be effectively applied in polymers, magnetic materials, and piezoelectric actuators.

Generalized rate-dependent hysteresis models provide a unified framework for describing, simulating, and predicting nonlinear memory effects in diverse physical, chemical, and engineering systems. These models systematically capture how the dissipation and history-dependent lag between input and output evolve as a function of the rate of external driving, enabling accurate prediction of behaviors such as strain stiffening, energy dissipation, loop scaling laws, and dynamic yield phenomena across materials including polymers, suspensions, magnetic solids, and piezoelectric actuators.

1. Mathematical Foundations and Core Model Classes

Generalized rate-dependent hysteresis models extend classic static frameworks by explicitly incorporating time derivatives and/or fractional derivatives into the constitutive relations between system state and input, thereby encoding both memory and rate dependence. Leading mathematical constructs include:

• Generalized Play and Nonlinear Play Operators: Classical Preisach/prandtl–ishlinskii models are static and encode no explicit time scale; generalized and nonlinear play operators introduce relaxation via differential inclusions of form

$$\frac{d}{dt}v(t) + C(\ell(u(t)), r(u(t)); v(t)) \ni 0,$$

where $v(t)$ is the internal state, and $C$ denotes a maximal-monotone “interval” operator (projecting $v$ onto the admissible range dictated by input $u$) (Peszynska et al., 2020). Assemblies of $K$ such “hysterons” allow rich, rate-dependent response, generalizing static relay-based models.

• Fractional-Order Models: The fractional-order normalized Bouc-Wen (FONBW) model introduces two non-integer-order derivatives in the evolution of the internal hysteretic variable:

$$\mathscr D^{\lambda_2} \bar{h}(t) = \rho \Big[\mathscr D^{\lambda_1} u(t) - \sigma |\mathscr D^{\lambda_1} u(t)| |\bar{h}(t)|^{n-1} \bar{h}(t) - (1-\sigma) \mathscr D^{\lambda_1} u(t) |\bar{h}(t)|^n\Big],$$

where $\lambda_1,\lambda_2\in(0,1]$ are fractional orders and $\mathscr D^\lambda$ denotes the Grünwald–Letnikov derivative. This formulation yields nonlocal-memory and robust rate effects (Kang et al., 2020).

• Dual Mechanism Constitutive Theories: Recent finite viscoelasticity models introduce two (or more) viscous "Maxwell" branches, each characterized by a distinct timescale and physical mechanism (e.g., molecular relaxation, glassy rearrangement). The total stress and Helmholtz free energy are additively decomposed:

$$\Psi(C;\,\{C^e_k, J^e_k\}) = \Psi_{\rm eq}(C,J) + \sum_{k=1}^{n} \Psi_{\rm neq}^{(k)}(C^e_k,J^e_k)$$

with flow evolution laws specified per mechanism, enabling transition from rubber-like to glassy/hardening response with increasing rate (Datta et al., 8 Jan 2026).

• Universal Data-Driven and Neural Network Models: Machine learning approaches (including Extended Preisach Neural Networks) can learn arbitrary rate-dependent hysteresis by mapping both instantaneous input $v(t)$0 and its derivative $v(t)$1 through layered architectures embedding deteriorating stop operators, sigmoidal nonlinearities, and universal approximation. This generalizes Preisach decomposition and allows for both congruent and non-congruent, symmetric and asymmetric, and rate-dependent hysteresis loops (Farrokh et al., 2019).

2. Universal Dynamical Scaling and Loop Area Dependence

A central result of generalized rate-dependent hysteresis theory is the emergence of universal power-law scaling for dissipated energy (loop area $v(t)$2) versus input sweep rate $v(t)$3. In particular, stochastic mean-field Langevin models for order parameters $v(t)$4 in a ramped field $v(t)$5, subject to thermal noise, yield (Sun et al., 25 Mar 2026):

• For $v(t)$6 (thermal regime), the excess area above quasi-static $v(t)$7 scales as $v(t)$8.
• For $v(t)$9 (athermal/dynamic regime), $C$0.

Experiments on magnetic materials, Ising simulations, and adsorption systems all follow this universal two-regime law. The crossover rate $C$1 is determined by system-specific parameters such as the critical temperature or activation energy and thus provides a direct design criterion for minimizing (or maximizing) hysteresis losses in devices.

3. Model Calibration, Parameter Identification, and Generalization

Practical application of generalized rate-dependent hysteresis models requires robust parameter identification, often accomplished via:

• Multi-branch fitting in dual-mechanism viscoelastic models: sequentially fitting quasi-static, moderate, and high-rate data to determine equilibrium and nonequilibrium parameters, e.g., loading–unloading loops and dynamic mechanical analysis for polymers (Datta et al., 8 Jan 2026).
• Evolutionary and optimization algorithms for inverse identification: self-adaptive differential evolution (for FONBW), genetic algorithms plus sub-gradient descent (for neural schemata), or automated symbolic regression (library-free, evolutionary search) to fit both structure and parameters to measured trajectories (Kang et al., 2020, Farrokh et al., 2019, Yang et al., 2 Dec 2025).
• Direct calibration of play/prandtl–ishlinskii-type discretizations from sparse primary and secondary curve data, leveraging hierarchical or adaptive partitioning for efficient numerical implementation (Peszynska et al., 2020).

Generalization is achieved by extending the number of internal mechanisms (Maxwell branches), accounting for additional dissipative effects (e.g., secondary β-mechanisms), or introducing internal structural variables describing state fraction, damage, or anisotropy.

4. Applications and Physical Interpretation Across Materials

Generalized rate-dependent hysteresis models have been validated in a wide array of domains:

• Elastomers and Glass-Formers: Dual-mechanism viscoelasticity captures both the classic Mullins-type rubber hysteresis at low rates and rate-induced glassy stiffening, yielding, and post-yield hardening at high rates, reproducing crossovers in dissipated energy and modulus frequency spectra (Datta et al., 8 Jan 2026).
• Soft Magnetic Cores and Adsorption Materials: Universal loop area scaling laws guide operational strategies to tune dynamic losses in transformers and gas-storage materials (Sun et al., 25 Mar 2026).
• Cohesive Suspensions and Yield-stress Fluids: Rate-dependent generalizations of the Herschel–Bulkley law, wherein the yield stress itself decays with rate (e.g., $C$2), explain non-monotonic flow curves, shear banding, and pronounced hysteresis/erratic yielding under controlled-stress driving. These models rationalize transitions from "cage" yielding at low Pe via bond-scale rupture at high Pe, and predict the dependence of yield strain on the characteristic rate (Buscall et al., 2014).
• Piezoelectric and Magnetostrictive Actuators: FONBW and data-driven neural network models deliver high-accuracy rate-dependent descriptions of actuator hysteresis under broad frequency sweeps, enabling precision tracking and compensation (Kang et al., 2020, Farrokh et al., 2019).
• Equation Discovery in Hysteretic Systems: Integration of differentiable hidden-variable solvers with symbolic regression yields governing equations directly from data, unifying the modeling of arbitrary, mechanism-unknown rate-dependent hysteresis (Yang et al., 2 Dec 2025).

5. Well-Posedness, Numerical Schemes, and Implementation

Robustness and computational tractability are secured via:

• Discrete resolvent and variational approaches: The discrete implicit update $C$3 for each play operator is directly implementable and ensures stability and monotonicity properties, aiding coupling to PDEs and large-scale simulations (with TV stability and first-order accuracy) (Peszynska et al., 2020).
• Fractional derivatives and history-summation: Practical implementation of fractional-order evolution requires efficient approximation schemes, e.g., Oustaloup filters or GL finite differences, to handle nonlocal memory effects (Kang et al., 2020).
• Modular integration in ODE/PDE solvers: Once closed-form or symbolic-regression equations are determined, integration with standard solvers (e.g., RK4, ode45, solve_ivp) is immediate, facilitating predictive simulation and control.

6. Classification, Model Selection, and Limitations

Model selection is governed by the application domain, available data, and desired interpretability:

Model Class	Key Features	Rate-Dependence Mechanism
Generalized play/nonlinear play	Piecewise/continuous loops	ODE inclusion, multiple time scales
Fractional Bouc-Wen (FONBW)	Asymmetry, memory, fractional	Fractional derivatives
Extended Preisach NN	Arbitrary loop shapes	Direct inclusion of $C$4 in network
Dual viscoelastic (Maxwell)	Multi-mechanism, glassy/rubbery	Explicit flow laws per branch

While all frameworks capture memory and rate effects, limitations include:

• Parameter identifiability may suffer if data coverage is sparse or limited to a narrow excitation range.
• Classical static Preisach and integer-order Bouc-Wen models do not suffice for strongly dynamic, frequency-dependent, or asymmetric systems.
• Black-box data-driven models may extrapolate poorly outside the regime of training unless appropriately regularized.

Generalized rate-dependent hysteresis frameworks systematically rationalize and predict complex experimentally observed phenomena, reconcile scattered empirical observations (such as diverse loop area power-law exponents), and support principled design of materials and devices exhibiting dynamic memory effects (Datta et al., 8 Jan 2026, Sun et al., 25 Mar 2026, Kang et al., 2020, Peszynska et al., 2020, Buscall et al., 2014, Yang et al., 2 Dec 2025, Farrokh et al., 2019).