The Hysteresis Paradox: When Systems Remember Their Past
Updated: Aug 21
Dr. Elena M. Kovács¹, Dr. Samuel R. Whitmore², Prof. Naoko Fujimura³
¹ Department of Nonlinear Materials Physics, Danube Institute of Advanced Science,
² Center for Complex Dynamical Systems, North Atlantic Technical University
³ Laboratory for Adaptive Matter and Memory, Kansai Frontier University
[Disclaimer: This is a sample academic article. All author names, affiliations, and institutional details are fictional and have been created solely for illustrative and educational purposes.]
Abstract
Hysteresis describes a fundamental class of history-dependent phenomena in which the present state of a system cannot be determined uniquely from its present external conditions because information about previous states has been retained within its internal configuration. In ferromagnetic materials, magnetization can persist after an applied magnetic field has been removed, producing characteristic hysteresis loops in which increasing and decreasing field trajectories follow different paths; phenomenological descriptions such as the Preisach model formalize this behavior through populations of switching elements with distinct transition thresholds [10.1007/BF01349418]. Comparable memory effects are produced in shape-memory alloys by reversible martensitic phase transformations, where thermal and mechanical histories determine phase fractions, strain, dissipation, and subsequent material response [10.1016/S0020-7462(02)00027-6; 10.1016/S1359-6454(99)00295-5]. In multiphase fluids and porous materials, drainage and imbibition can follow different pressure–saturation pathways because interfaces become pinned, depinned, and reorganized within complex pore geometries, causing macroscopic fluid states to depend on the route through which they were prepared [10.1103/PhysRevE.73.016307; 10.1029/2003WR002908]. At larger scales, positive feedback and alternative stable states can produce ecological regime shifts in which reversal of the original forcing does not immediately restore the previous ecosystem state, demonstrating that hysteresis can arise far beyond conventional materials physics [10.1038/35098000]. Such phenomena are commonly associated with metastability, energy barriers, cooperative switching, nonlinear feedback, disorder, and multiple accessible states, through which microscopic interactions are converted into macroscopic memory. Hysteresis therefore presents an apparent paradox: deterministic physical laws may govern individual interactions, yet knowledge of the instantaneous external input alone remains insufficient for predicting the system output because part of the relevant state information is encoded in its trajectory through state space. This article examines the physical origins of hysteresis, including magnetic domain switching, metastable energy landscapes, martensitic transformations, capillary pinning, nonlinear feedback, return-point memory, noise-driven transitions, and multistability, while considering applications in nonvolatile memory, sensing, adaptive materials, actuators, neuromorphic devices, robotics, and environmental modeling. Particular emphasis is placed on distinguishing genuine physical memory from simple delayed response and on explaining why hysteresis can simultaneously represent energy dissipation, information storage, irreversibility, and a useful engineering functionality. The resulting framework suggests that many complex systems should not be represented solely by the conventional relation output = function(current input), but instead by the more general principle output = function(current input, internal state, and history).
1. Introduction
Hysteresis arises when the state reached by a system depends not only on the value of the present driving variable but also on the sequence of states through which the system has previously passed, making history an effective state variable. Although the phenomenon was first formalized extensively in magnetism, closely related behavior is encountered in martensitic phase transformations, capillary systems, frictional and mechanical structures, electronic memory devices, ecosystems, and climate models [10.1016/0304-8853(86)90066-1; 10.1016/S0020-7462(02)00027-6; 10.1103/PhysRevE.73.016307]. A characteristic hysteresis loop is obtained when an external control parameter is increased and subsequently decreased while the measured output follows different trajectories, demonstrating that the mapping between input and output is not single-valued unless information about internal state or previous reversal points is retained [10.3390/ma16062422]. Such behavior is commonly associated with energy barriers, metastable states, cooperative switching, pinning, structural rearrangement, or feedback, all of which can prevent a system from instantaneously reaching the state that would minimize its equilibrium free energy under current conditions [10.1103/PhysRevLett.70.3347; 10.1016/S0020-7462(02)00027-6]. Hysteresis should therefore be distinguished from a simple time delay: a delayed linear system may eventually reach the same equilibrium state after sufficient waiting, whereas a genuinely hysteretic system can retain different states at the same control parameter even in the quasistatic limit [10.1016/S0020-7462(02)00032-X]. The resulting “hysteresis paradox” is that apparently identical present conditions can generate different outputs without violating deterministic physical law, because the missing information is encoded in hidden internal variables and in the path by which the system reached its present configuration.
2. Results and Discussion
The central feature of hysteresis can be expressed schematically as (y=\mathcal{H}[x(t)]), where the output (y) depends on the history of the input trajectory rather than only on its instantaneous value (x(t)). This differs fundamentally from an ordinary constitutive relation (y=f(x)), because two trajectories arriving at the same (x) may correspond to different populations of magnetic domains, martensitic variants, fluid interfaces, ionic configurations, ecological states, or other internal structures [10.1016/j.ymssp.2014.04.012]. Hysteresis models must therefore incorporate internal state, switching thresholds, reversal points, or distributed metastable elements capable of retaining information about earlier forcing [10.3390/ma16062422]. The scientific problem is consequently not merely to identify a loop in an input–output graph, but to determine which microscopic mechanisms store the information represented by that loop and under what conditions this memory is retained, erased, or reorganized.
2.1. Hysteresis Is Memory, Not Merely Slow Response
The term hysteresis is sometimes used loosely for any situation in which an output lags behind an input, but rate-independent hysteresis represents a more specific form of memory in which different outputs can persist for the same input even as the driving rate is made arbitrarily small [10.1016/S0020-7462(02)00032-X]. In a simple relaxation process, reducing the rate of forcing allows the system to approach a unique equilibrium curve and the apparent loop progressively disappears, whereas a rate-independent hysteretic system retains distinct increasing and decreasing branches because transitions are controlled by internal thresholds or metastable configurations [10.3390/en16093908]. This distinction is important experimentally because dynamic lag caused by viscosity, inertia, diffusion, or finite reaction kinetics can resemble hysteresis when measurements are performed too rapidly. A rigorous analysis therefore requires tests over multiple driving rates together with characterization of reversal curves and internal state variables so that true history-dependent memory can be separated from ordinary transient response [10.1016/j.ymssp.2014.04.012]. Hysteresis is thus most usefully interpreted as a constitutive memory of the path rather than simply as delayed tracking of a changing stimulus.
2.2. Magnetic Hysteresis: Domains Store the History of the Applied Field
Ferromagnetism provides the canonical example of hysteresis because magnetization depends strongly on how magnetic domains and domain walls have previously responded to an applied magnetic field. As the field is increased, domain walls move, domains favorably aligned with the field grow, and magnetic moments rotate toward the field direction, but defects, grain boundaries, internal stresses, and other microstructural features can impede or pin these processes [10.1016/0304-8853(86)90066-1]. When the applied field is subsequently reduced to zero, the microscopic domain configuration generally does not return to its original state, leaving a remanent magnetization; an oppositely directed field must then be applied to reach zero net magnetization, defining the coercive field [10.1016/0304-8853(86)90066-1]. The closed (M-H) or (B-H) loop records this history, and its area is related to irreversible energy dissipation during a magnetization cycle. Models such as the Jiles–Atherton framework connect macroscopic hysteresis to irreversible domain-wall motion and pinning, whereas Preisach-type descriptions represent the material as an ensemble of bistable switching units with different thresholds [10.1016/0304-8853(86)90066-1; 10.3390/ma16062422]. Magnetic memory therefore emerges because microscopic switching events are not completely reversed when the external field retraces its path.
2.3. Metastability and Energy Barriers: Why Systems Do Not Immediately Find the Lowest-Energy State
Many hysteretic systems can be understood through an energy landscape containing several local minima separated by barriers, where each minimum corresponds to a metastable configuration that may persist even when another configuration has become thermodynamically more favorable. Gradually changing an external parameter deforms this landscape, but the system can remain trapped in its existing basin until the associated barrier disappears or a sufficiently large fluctuation drives a transition, at which point an abrupt jump to another state occurs [10.1103/PhysRevLett.70.3347]. When the control parameter is reversed, the corresponding reverse transition generally occurs at a different threshold because the system now occupies a different local minimum, producing separate forward and backward branches. Disorder can broaden the distribution of switching thresholds and cause many microscopic transitions to occur as intermittent avalanches rather than as a single smooth transformation [10.1103/PhysRevLett.70.3347]. Hysteresis is therefore intimately connected with metastability: the present configuration may not represent the globally lowest-energy state but can remain dynamically stable because an energy barrier preserves information about how the state was created.
2.4. Return-Point Memory: A Hysteretic System Can Remember More Than One Turning Point
Some hysteretic systems possess a remarkable property known as return-point memory, in which reversing the external drive at a particular value, executing a partial loop, and later returning to that value can restore the system to essentially the same macroscopic and sometimes microscopic state that existed at the previous visit [10.1103/PhysRevLett.70.3347]. The random-field Ising model provided a theoretical demonstration that disorder-driven hysteretic systems can exhibit nested subloops and a hierarchy of remembered reversal points under appropriate ordering conditions [10.1103/PhysRevLett.70.3347]. Related return-point memory has been studied in deterministic cellular automata and artificial spin-ice systems, demonstrating that sophisticated memory behavior can emerge from large collections of interacting or threshold-switching elements without a conventional digital memory register [10.1103/PhysRevLett.72.2203; 10.1103/PhysRevE.86.021406]. Preisach-type models similarly possess a “wiping-out” property in which sufficiently large excursions erase selected portions of previous input history while more recent reversal points remain encoded [10.3390/ma16062422]. Hysteresis can therefore provide structured rather than indiscriminate memory: the system may compress its past into an ordered sequence of physically significant turning points.
2.5. Shape-Memory Alloys: Structural History Becomes Mechanical Function
Shape-memory alloys such as NiTi exploit a hysteretic martensitic transformation in which crystal structure, temperature, stress, and previous deformation history jointly determine the macroscopic shape and mechanical response of the material [10.1016/S1359-6454(99)00295-5]. Cooling can drive a transformation from the high-temperature austenite structure into martensitic variants, while heating reverses the transformation, but the temperatures at which forward and reverse transformations occur are separated by a hysteresis interval because nucleation, interface motion, compatibility constraints, and dissipative processes impede perfectly reversible transformation [10.1016/S0020-7462(02)00027-6]. Under mechanical loading, stress-induced martensite can similarly produce large recoverable strains and superelastic behavior, with the loading and unloading curves forming characteristic hysteresis loops [10.1016/S1359-6454(99)00295-5]. Thermomechanical training can modify internal microstructures and transformation pathways, meaning that repeated history can influence subsequent actuation behavior. In this case, hysteresis is not merely a limitation but a central component of functionality, enabling thermally activated actuators, adaptive structures, deployable systems, and mechanically responsive devices whose behavior is determined by stored structural history [10.1016/S0020-7462(02)00027-6].
2.6. Fluids and Porous Media: Interfaces Remember Whether the System Was Filled or Drained
Hysteresis also occurs in fluid systems in which moving interfaces encounter complex geometry, surface-energy barriers, contact-line pinning, and disconnected pore networks. In unsaturated porous media, the relationship between capillary pressure and liquid saturation differs during drainage and imbibition because liquid withdrawal is controlled largely by pore throats, whereas refilling depends on access to larger pore bodies, contact angles, and trapping of fluid phases [10.1029/2003WR002908]. A given capillary pressure can therefore correspond to different water saturations depending on whether the porous material has previously been drying or wetting [10.1029/2003WR002908]. Macroscopic theories in which residual fluid contents are introduced as additional state variables can reproduce process dependence and hysteretic capillary-pressure relations, illustrating that the apparent nonuniqueness disappears once sufficient internal-state information is supplied [10.1103/PhysRevE.73.016307]. Similar principles occur in droplets, wetting, adsorption, boiling, microfluidics, and multiphase transport, where pinned interfaces can preserve geometric information about earlier conditions. Fluid hysteresis thus provides a clear example of memory being stored not electronically or magnetically but in the spatial configuration of interfaces.
2.7. Environmental Hysteresis: Returning the Forcing Does Not Necessarily Restore the Original State
History dependence can also emerge at ecosystem and climate scales when positive feedback creates multiple stable or metastable macroscopic states. Ecological systems can undergo transitions between alternative regimes, such as clear and turbid states in shallow lakes, and reversal of the environmental forcing may require conditions substantially different from those that originally triggered the transition because feedback stabilizes the newly established regime [10.1038/35098000]. Consequently, restoring nutrient concentration, harvesting intensity, or another driver to its value immediately before collapse may be insufficient to restore the previous ecosystem state [10.1038/35098000]. Climate models provide another example: intercomparison experiments in which freshwater forcing of the North Atlantic was increased and subsequently reduced produced hysteresis in thermohaline circulation across multiple models, with strong and weak circulation states existing over overlapping ranges of forcing [10.1029/2005GL023655]. These results should not be interpreted as evidence that every environmental variable necessarily exhibits a fixed irreversible threshold, but they demonstrate that nonlinear feedback can make environmental recovery strongly path-dependent. Management strategies based solely on restoring present external conditions can therefore fail when the internal state of the system has already crossed into a different basin of attraction.
2.8. Noise, Disorder, and Avalanches: Memory Can Be Written Through Collective Switching
Real hysteretic systems are rarely perfectly homogeneous, and microscopic disorder creates distributions of local transition thresholds that can strongly shape macroscopic response. In disorder-driven models of first-order transitions, gradual changes in an external field cause local regions to switch individually or cooperatively, producing avalanches whose sizes can range from microscopic events to abrupt system-spanning transformations [10.1103/PhysRevLett.70.3347]. Noise can assist transitions across energy barriers, erase weakly stabilized memories, or alter the precise sequence of switching events, while stronger disorder may fragment a large transition into many smaller events. Counterintuitively, disorder can sometimes improve aspects of memory by pinning defects and preventing uncontrolled rearrangement, as observed in simulations of return-point memory in artificial spin-ice systems [10.1103/PhysRevE.86.021406]. The macroscopic hysteresis loop therefore represents an aggregate record of numerous microscopic events whose exact sequence may be partially stochastic while their statistical organization remains reproducible.
2.9. Memristive Electronics: Electrical Resistance That Depends on Its Own History
Hysteretic memory has become technologically important in nanoscale electronics through memristive devices, whose resistance depends on internal variables modified by the history of applied voltage or current. A physical model of a titanium-dioxide memristive device showed that coupled electronic and ionic transport can generate a characteristic pinched current–voltage hysteresis loop, with movement of charged species modifying the internal resistance state [10.1038/nature06932]. Because the altered internal configuration can persist after the driving voltage is removed, information can be stored nonvolatily in the device [10.1038/nature06932]. Subsequent development of memristive technologies has targeted nonvolatile memory, analog computation, in-memory processing, and neuromorphic systems in which gradual history-dependent conductance changes can emulate aspects of synaptic plasticity [10.1038/nnano.2012.240]. In this context, hysteresis becomes an engineered information-storage mechanism: instead of eliminating path dependence as an undesirable nonlinearity, device operation is specifically designed so that electrical history modifies the future response.
2.10. Hysteresis in Sensors, Actuators, and Robotics
In precision engineering, hysteresis can either degrade performance or provide useful mechanical intelligence depending on how it is exploited. Piezoelectric, magnetic, and shape-memory actuators can exhibit different displacements under increasing and decreasing drive, generating positioning errors if they are controlled using a single-valued calibration curve [10.1016/j.ymssp.2014.04.012]. Hysteresis compensation is therefore frequently implemented through inverse Preisach, Prandtl–Ishlinskii, Bouc–Wen, or related models that estimate the internal history state and calculate an input capable of producing the desired output [10.1016/j.ymssp.2014.04.012]. Conversely, shape-memory structures and memristive control elements can intentionally exploit hysteretic transitions to produce mechanical latching, persistent configurations, threshold-triggered actuation, or behavior that depends on previous stimulation [10.1016/S0020-7462(02)00027-6; 10.1038/nnano.2012.240]. This suggests a broader engineering shift in which memory need not always be implemented in a separate digital controller; it can be physically embedded into the material or device response itself.
2.11. Modeling a System That Contains Its Own History
A fundamental difficulty in modeling hysteresis is that an instantaneous pair of input and output values does not uniquely specify the future trajectory, so additional variables representing the internal memory state must be introduced. Preisach models represent hysteresis using a distribution of elementary two-state operators with different switching thresholds, while Prandtl–Ishlinskii models use combinations of play or stop operators and Bouc–Wen-type models describe hysteresis through internal differential variables [10.3390/ma16062422; 10.1016/j.ymssp.2014.04.012]. These models differ in physical interpretation and mathematical convenience, but all recognize that the state of a hysteretic system contains information not visible in the instantaneous external input. Data-driven models can supplement such frameworks when microscopic state variables cannot be measured directly, although prediction outside the range of previously observed histories remains challenging. A robust model should therefore be tested not merely on major hysteresis loops but on partial loops, nested reversals, rate changes, repeated cycling, and previously unseen paths if genuine predictive memory is to be demonstrated [10.1016/j.ymssp.2014.04.012].
2.12. Microscopic Reversibility and Macroscopic Irreversibility
The existence of hysteresis does not require a violation of microscopic physical laws even when those laws are approximately invariant under time reversal. Macroscopic irreversibility can emerge because enormous numbers of microscopic degrees of freedom are not individually tracked, energy is transferred into internal vibrations or heat, transitions occur through metastable states, and coarse-grained descriptions discard information required to reconstruct the exact microscopic trajectory. In ferromagnets, irreversible domain-wall jumps and pinning generate dissipative macroscopic loops, while in shape-memory alloys the movement of phase boundaries and structural rearrangements similarly dissipate energy during transformation [10.1016/0304-8853(86)90066-1; 10.1016/S0020-7462(02)00027-6]. The apparent paradox is therefore resolved by recognizing that a macroscopic state variable such as magnetization, strain, saturation, or ecosystem biomass represents only a tiny subset of the full microscopic state. History dependence appears because hidden variables retain information that has been removed from the reduced description.
3. Conclusion and Outlook
Hysteresis demonstrates that the present state of a system can contain a physically encoded record of its past, invalidating any model in which output is assumed to depend only on the instantaneous value of an external input. In ferromagnets, memory is stored through domain configurations and pinning [10.1016/0304-8853(86)90066-1]; in shape-memory alloys it is stored through martensitic microstructure and transformation history [10.1016/S0020-7462(02)00027-6]; in porous media it is stored in interfacial configurations and trapped phases [10.1103/PhysRevE.73.016307]; and in memristive electronics it is stored through persistent changes in ionic and electronic configuration [10.1038/nature06932]. Return-point memory further demonstrates that hysteretic systems can encode nested sequences of previous reversal points rather than merely a single binary past state [10.1103/PhysRevLett.70.3347]. At ecosystem and climate scales, feedback-driven multistability shows why reversing an external forcing may not necessarily retrace the original trajectory [10.1038/35098000; 10.1029/2005GL023655]. Future research is increasingly likely to move from merely characterizing hysteresis as an unwanted deviation from ideal behavior toward deliberately engineering physical memory, in which materials and devices retain useful information through their internal states. The resulting conceptual progression can be written as input history → internal-state evolution → metastability and switching → present response → stored memory → path-dependent future behavior, suggesting that some of the most adaptive future materials, sensors, robots, and computing systems may be designed not to forget their history, but to use it.
References
Jiles, D. C., & Atherton, D. L. (1986). Theory of ferromagnetic hysteresis. Journal of Magnetism and Magnetic Materials, 61, 48–60. [10.1016/0304-8853(86)90066-1]
Sethna, J. P., Dahmen, K., Kartha, S., Krumhansl, J. A., Roberts, B. W., & Shore, J. D. (1993). Hysteresis and hierarchies: Dynamics of disorder-driven first-order phase transformations. Physical Review Letters, 70, 3347–3350. [10.1103/PhysRevLett.70.3347]
Goicoechea, J., & Ortín, J. (1994). Hysteresis and return-point memory in deterministic cellular automata. Physical Review Letters, 72, 2203–2206. [10.1103/PhysRevLett.72.2203]
James, R. D., & Hane, K. F. (2000). Martensitic transformations and shape-memory materials. Acta Materialia, 48, 197–222. [10.1016/S1359-6454(99)00295-5]
Ortín, J. (2002). Hysteresis in shape-memory alloys. International Journal of Non-Linear Mechanics, 37, 1275–1281. [10.1016/S0020-7462(02)00027-6]
Visintin, A. (2002). On hysteresis in elasto-plasticity and in ferromagnetism. International Journal of Non-Linear Mechanics, 37, 1283–1298. [10.1016/S0020-7462(02)00032-X]
Scheffer, M., Carpenter, S., Foley, J. A., Folke, C., & Walker, B. (2001). Catastrophic shifts in ecosystems. Nature, 413, 591–596. [10.1038/35098000]
Tokunaga, T. K., Olson, K. R., & Wan, J. (2004). Conditions necessary for capillary hysteresis in porous media: Tests of grain size and surface tension influences. Water Resources Research, 40, W05111. [10.1029/2003WR002908]
Rahmstorf, S., Crucifix, M., Ganopolski, A., et al. (2005). Thermohaline circulation hysteresis: A model intercomparison. Geophysical Research Letters, 32, L23605. [10.1029/2005GL023655]
Hilfer, R. (2006). Macroscopic capillarity and hysteresis for flow in porous media. Physical Review E, 73, 016307. [10.1103/PhysRevE.73.016307]
Strukov, D. B., Snider, G. S., Stewart, D. R., & Williams, R. S. (2008). The missing memristor found. Nature, 453, 80–83. [10.1038/nature06932]
Libál, A., Reichhardt, C., & Olson Reichhardt, C. J. (2012). Hysteresis and return-point memory in colloidal artificial spin ice systems. Physical Review E, 86, 021406. [10.1103/PhysRevE.86.021406]
Yang, J. J., Strukov, D. B., & Stewart, D. R. (2013). Memristive devices for computing. Nature Nanotechnology, 8, 13–24. [10.1038/nnano.2012.240]
Hassani, V., Tjahjowidodo, T., & Do, T. N. (2014). A survey on hysteresis modeling, identification and control. Mechanical Systems and Signal Processing, 49, 209–233. [10.1016/j.ymssp.2014.04.012]
Mörée, G., & Leijon, M. (2023). Review of Play and Preisach models for hysteresis in magnetic materials. Materials, 16, 2422. [10.3390/ma16062422]
Mörée, G., & Leijon, M. (2023). Review of hysteresis models for magnetic materials. Energies, 16, 3908. [10.3390/en16093908]
Comments