Skip to main content
computer-science economics engineering-and-technology medicine-and-health physics psychology

​
Hysteresis

​
Description

A system exhibits hysteresis when its current output depends not just on the current input but on the path of past inputs — the history sticks around. The same input can produce different outputs depending on what came before. Classic cases: magnetic-material hysteresis loops, thermostat dead-bands (heating on at 68°F but off at 72°F), biological refractory periods (a neuron that just fired can’t fire again immediately). The diagnostic question — “does the system give the same output for the same input regardless of history, or does the path matter?” — distinguishes hysteresis from fixed-point behavior. Where hysteresis is present, stabilizing systems often deliberately use it (dead-bands prevent oscillation around the setpoint); destabilizing systems often suffer from it (once-recovered behavior depends on which crisis-path got you there).

​
Triggers

User-initiated: User describes a system whose behavior “remembers” past states, where recovery takes a different path than degradation, or where a small disturbance has a disproportionate lingering effect. Vocabulary cues: “path dependence,” “sticky,” “lag,” “remember,” “lock-in.” Agent-initiated: Agent notices that the same input is producing different outputs at different times, and the only explanation is the system’s recent history. Candidate inference: “the path matters; what state from the past is still active?” Situation-shape signals: Loops that don’t close cleanly. Recovery curves that don’t mirror degradation curves. A/B testing where the order of conditions matters more than expected. Configuration changes that don’t fully reverse on rollback.

​
Exclusions

  • Memoryless / Markovian systems — the future depends only on current state, not on history. Many ideal models assume this; reality often violates it.
  • Pure stateless functions — f(x) returns the same output for the same input every time, regardless of history. The concept doesn’t fire.
  • Hysteresis-as-bug rather than hysteresis-as-concept — sometimes “the output depended on history” is a bug we should design out, not a feature we should name. Distinguish: was the hysteresis intentional (engineered) or accidental (leaked from below)?

​
Structure

Internal structure of hysteresis: a table of its component slots and the concepts that fill them.

​
Relationships

Relationship neighborhood of hysteresis: a graph of the concepts it connects to and the concepts it is a part of.
  • backpressure — backpressure systems often exhibit hysteresis on the recovery side; once throttled, return to full throughput depends on the path taken.
  • graduation-promotion — promotion has hysteresis built in; the cost of demotion is asymmetric to the cost of promotion.
  • asymmetric-gate — hysteresis is a path-asymmetric form of the asymmetric-gate; cheap to enter, expensive to leave (or vice versa).
  • active-gate-vs-passive-audit — dead-band gates use deliberate hysteresis to prevent flutter at the gate boundary.
  • one-way-ratchet — one-way-ratchet is hysteresis taken to the limit: the reverse path is structurally absent.

​
Examples

Thermostat dead-band · engineering-and-technology

engineered hysteresis prevents rapid on-off cycling; the system “remembers” whether it’s currently heating.

Lock-in / switching costs · economics

once committed to a platform or vendor, the cost of switching is asymmetric to the cost of staying; the choice persists.
Status-quo bias and the endowment effect are hysteresis in human preference. Standard rational-choice theory is memoryless: which option you pick should depend only on the options’ merits, not on which one you happen to be holding. The behavioral evidence says otherwise. Samuelson and Zeckhauser (1988) showed that an option is chosen far more often when it is labeled the current state than when presented neutrally — people stick. Kahneman, Knetsch, and Thaler’s mug experiments (1990) showed the same asymmetry from the other side: people randomly given a mug demanded roughly twice as much to give it up (willingness-to-accept) as people not given one were willing to pay to acquire it (willingness-to-pay). The current valuation depends on the prior endowment, not just on the object.This is the defining hysteresis signature: the system’s present state depends on the path it took to get there, not only on present conditions. The mechanism is loss aversion — losses loom larger than equivalent gains — which puts a kink in the utility function at the current reference point. Moving away from wherever you are is coded as a loss, so you resist; the reference point is the “memory” the system carries, exactly analogous to a magnetic material that depends on which direction it was last magnetized or a thermostat that depends on which side of the dead-band it last left from.Inference: When you find a preference that depends on the starting allocation rather than just the options, you are in a hysteretic regime, and interventions that assume memoryless choice will misfire. The famous corollary is the violation of the Coase theorem: if the final allocation depends on the initial assignment of rights, “the market will reallocate to the efficient outcome regardless of who starts with what” fails. Defaults become powerful policy levers precisely because of this hysteresis — opt-out organ donation and auto-enrollment retirement plans work by setting the reference point, since people stay where they are placed.
long-running internal states whose current behavior depends on the path of past consumption.
In a ferromagnetic material, the magnetization at a given applied field depends not just on the current field but on the history of fields the material has been exposed to. Plotting magnetization against applied field over a cycle of increasing and decreasing field produces a closed loop — the hysteresis loop — rather than a single curve. James Ewing studied and named “hysteresis” in this physics context in the late 19th century.The structural property the loop visualizes is path-dependence: the system’s current state retains information about its past, so two histories that arrive at the same applied-field input can leave the material in different magnetic states. This is the original empirical case from which the concept gets its name, and the loop diagram is the canonical pictorial representation of hysteresis across other domains that borrow the term.
the original physics case; magnetization depends on past applied field.
behavioral hysteresis; once a user has experienced a failure, their behavior depends on having done so even after conditions revert.
Paul David’s 1985 paper “Clio and the Economics of QWERTY” argued that the now-ubiquitous QWERTY keyboard layout is not the result of an efficient market choosing the optimal arrangement, but the path-dependent outcome of early-twentieth-century mechanical-typewriter constraints, training capital, and complementary investments that locked in a layout once it crossed a critical mass. Brian Arthur’s 1989 work on Competing Technologies, Increasing Returns, and Lock-In by Historical Small Events formalized the dynamics: with increasing returns, small early advantages amplify into permanent market positions, and the order in which competing alternatives arrive shapes the eventual equilibrium more than their intrinsic merits.Inference: The economic path-dependence literature is the social-science instantiation of hysteresis at scale. The system’s current state (which keyboard layout is universal; which video format wins; which programming language captures a generation) depends on the historical path, not just on present-day input. The corollary is that interventions designed assuming Markovian behavior fail in path-dependent regimes — policymakers who assume “the market will correct toward the optimum” miss that the market is sitting in a deep basin shaped by the past and cannot reach a better basin without an explicit unlock move. Adoption decisions in network-effect industries inherit this directly: timing and seeding matter more than ultimate quality once lock-in has taken hold.
James Alfred Ewing coined the term hysteresis (from the Greek husteresis, “a coming short”) in 1881 to describe the magnetic-material behavior in which the magnetization of iron does not retrace the same curve when the external field is reduced as it traced when the field was applied. The magnetization-vs-field plot becomes a closed loop — the “hysteresis loop” — and the area inside the loop is the energy dissipated per cycle. The same mathematical structure later turned out to govern thermostat behavior (the heater switches on at a low setpoint and off at a higher one, with the dead-band preventing oscillation), biological refractory periods (a neuron just-fired cannot fire again for a few milliseconds regardless of input), and a long list of engineered systems that deliberately use hysteresis for stability.Inference: The cross-domain published lineage spanning physics, engineering, biology, and economics is exactly the field-validated character that earns the concept’s place in the catalog. The Ewing→thermostat→neuron→QWERTY transfer is not metaphor; it is shared mathematical structure (output as a function of input and integrated path history) showing up in different substrates. Once the pattern is recognized, the design lesson follows: where stability matters, intentional hysteresis (Schmitt triggers, dead-bands, debouncing) is a cheap insurance against the chatter that otherwise burns lifetime; where corrigibility matters, accidental hysteresis is the warning sign that a system you wanted to be Markovian is silently accumulating state.
service degradation engages, throttling activates, but full recovery requires sustained good-input rather than just current good-input.
A simple thermostat is the textbook example of engineered hysteresis. Rather than turning the heat on the instant the temperature drops below a setpoint and off the instant it rises above it — which would produce rapid on-off cycling around the setpoint — the controller is designed with a “dead-band”: heat turns on at the lower threshold (say 68°F) and off at the upper threshold (say 72°F). The system’s current output depends not just on the current temperature but on whether it is in the heating phase or the not-heating phase.The same shape recurs as the Schmitt trigger in electronics and as dead-band controllers more generally: building hysteresis into the loop deliberately to suppress chattering near the threshold. The reading reveals that hysteresis can be a design choice (preventing rapid cycling, providing memory of which regime is active) rather than only an undesired artifact.Inference: When a system is chattering between two states near a threshold, adding hysteresis — separating the trigger-on and trigger-off thresholds — is often the simplest stabilization. The cost is that the system carries a memory of which side it last left from, which has to be accounted for downstream.