Hysteresis — What Systems Remember

A thermostat in a house turns on the heat at sixty-eight degrees and turns it off at seventy-two. If the temperature hovers around seventy, the furnace cycles on and off rather than settling at a single point. The system remembers which direction the temperature came from. It responds differently to a room that is warming up than to one that is cooling down, even when the temperature is the same.

This is hysteresis. The term comes from the Greek word for lagging or lagging behind, and it describes a general phenomenon: the state of a system at any moment depends not only on the current inputs but on the history of those inputs. The system carries a memory of what happened before, and that memory changes how it responds to what happens now.

The term was coined in 1881 by James Alfred Ewing, a Scottish physicist studying magnetic materials. He observed that when a magnetic material is exposed to a magnetic field, it does not return to its original state when the field is removed. The magnetization lags behind the applied field. If you plot magnetization against field strength, the curve traces one path as the field increases and a different path as it decreases. The loop formed by the two paths is called a hysteresis loop.

The phenomenon appears in many fields under different names and in different contexts. The underlying structure is the same: a system that does not simply mirror its inputs but responds based on where it has been.

The magnetic origin

The original observation came from studying ferromagnetic materials — iron, nickel, cobalt, and their alloys. When you apply a magnetic field to a piece of iron, the magnetic domains inside the material align with the field. When you remove the field, some of that alignment remains. The material is now magnetized even though no external field is acting on it. This remaining magnetization is called remanence.

To demagnetize the material, you must apply a field in the opposite direction. The strength of that reverse field required to bring the magnetization to zero is called the coercivity. The combination of remanence and coercivity determines the shape of the hysteresis loop.

The loop is not a quirk of iron. It is a structural feature of any system with internal states that change slowly relative to the inputs. The domains inside the material do not rearrange instantly. They get stuck on defects in the crystal lattice. They require a threshold force to move. Once they move, they do not immediately move back.

This is the essential mechanism of hysteresis: the system has internal states that change with some resistance, and that resistance creates a gap between what the system is experiencing now and what it will respond to.

The engineering perspective

In engineering, hysteresis is often treated as a nuisance. It introduces error. A sensor that reads differently depending on whether the measured quantity is increasing or decreasing is less useful than one that responds the same way in both directions. Engineers spend significant effort designing systems that minimize hysteresis — precision instruments, measurement devices, control systems that require accuracy.

But hysteresis is also useful. The thermostat example above is one case. A thermostat without hysteresis would turn the heat on and off every time the temperature crossed seventy degrees. In practice, temperature fluctuates. The furnace would cycle rapidly, wasting energy and wearing out the equipment. The hysteresis band — the gap between the on and off thresholds — prevents this rapid cycling. It introduces a deliberate lag that stabilizes the system.

Hysteresis is built into relay controllers, which switch between states at different input levels depending on the current state. A relay that turns a motor on at fifty volts and off at forty-five volts behaves differently from a relay that turns on and off at the same threshold. The first resists rapid switching. The second chatters when the voltage is near the threshold. The hysteresis band is not a design flaw. It is a feature that prevents unwanted oscillation.

The same principle appears in mechanical systems. A spring that stretches differently depending on whether it is being loaded or unloaded exhibits hysteresis. The energy lost in the hysteresis loop is dissipated as heat. This is why hysteresis dampers are used in earthquake-resistant buildings. The material absorbs energy from the building’s motion and dissipates it, reducing the amplitude of the oscillation.

The economic lesson

Hysteresis entered economic discourse in the late twentieth century, most prominently through the work of George Akerlof and Janet Yellen, who edited the 1991 volume Hysteresis and the Economics of Unemployment. The central claim was simple but controversial: a recession is not just a temporary deviation from a stable equilibrium. It can change the equilibrium itself.

In standard macroeconomic models, unemployment has a natural rate — the level consistent with stable inflation. When unemployment rises above this rate due to a negative demand shock, the model predicts that wages will eventually fall, prices will adjust, and the economy will return to the natural rate. The shock leaves a temporary scar.

Hysteresis says the scar can be permanent. If unemployment stays high for a long enough period, workers lose skills. They become disconnected from the labor market. Employers begin to view them as unemployable. The natural rate itself rises. When the recovery comes, it does not return to the original unemployment level. It settles at a higher one.

The mechanism is straightforward. Long-term unemployment erodes human capital. Workers who are out of the labor market for months or years forget how to search effectively. Their skills atrophy. Their confidence declines. Employers discriminate against them, interpreting the unemployment spell as a signal of low ability even when the cause was a recession. The result is that the economy’s capacity to employ people has shrunk.

This is not a claim that all recessions cause permanent damage. It is a claim that the damage depends on duration. A short recession may leave little trace. A prolonged one can shift the entire distribution of employment. The system remembers the shock.

The policy implication is significant. If hysteresis is real, then the cost of a recession is not just the output lost during the downturn. It is also the permanent reduction in the economy’s productive capacity. The argument for early intervention — for fiscal stimulus, for monetary easing, for unemployment insurance that prevents skill loss — becomes stronger. The economy does not automatically heal. It requires active measures to return to the pre-shock equilibrium.

Whether hysteresis actually operates in modern economies is debated. Some economists argue that labor markets are flexible enough to prevent permanent damage. Others point to the elevated unemployment rates in Europe during the 1980s and 1990s as evidence that recessions can have lasting effects. The International Monetary Fund has acknowledged the possibility in multiple publications, noting that deep and prolonged downturns can raise the unemployment rate through mechanisms like skill erosion and insider-outsider dynamics.

The ecological dimension

Ecology has long recognized hysteresis, though the term is less commonly used than the phenomenon itself. A lake that becomes eutrophic — overloaded with nutrients, dominated by algae — does not simply clear up when the nutrient input is reduced. The sediment at the bottom of the lake continues to release phosphorus. The algae-dominated state is self-reinforcing. It takes a much larger reduction in nutrients to restore the lake than the amount that caused the shift in the first place.

This is the same structural pattern as the magnetic hysteresis loop. The lake’s state (clar vs. algae-dominated) depends not only on the current nutrient level but on the history of nutrient inputs. The system has two stable states at the same input level. Which one it occupies depends on which one it was already in.

Similar patterns appear in other ecosystems. A coral reef that shifts to an algae-dominated state does not revert when water quality improves. A grassland that converts to shrubland does not recover when rainfall returns. The shift is often abrupt — the system appears stable until a threshold is crossed, at which point it collapses into a different configuration. Reversing the collapse requires pushing the system past a different threshold in the opposite direction.

The practical implication is that restoring a degraded ecosystem is harder than preventing the degradation in the first place. The amount of intervention required to return a system to its original state exceeds the amount of pressure that caused the shift. The system’s memory of its degraded state is encoded in the internal dynamics — the nutrient cycling, the species interactions, the feedback loops — that sustain the new configuration.

What hysteresis reveals

The common thread across all these domains is that systems are not simple mirrors of their inputs. They have internal structure — domains in a magnet, skills in a worker, species in a lake — that changes slowly and resists reversal. This structure creates a gap between the input that caused a change and the input required to undo it.

This gap has several important consequences.

First, the path taken to reach a state matters. Two systems at the same input level may be in different states depending on whether they arrived from above or below. The history is encoded in the current configuration.

Second, reversing a change is often harder than causing it. The threshold for reversal is different from the threshold for the initial shift. This means that prevention is easier than restoration, whether the system is economic, ecological, or mechanical.

Third, the concept of equilibrium becomes more complicated. A system with hysteresis does not have a single equilibrium for each input level. It can occupy multiple states at the same input level, and which state it occupies depends on history. The mapping from input to state is not a function in the mathematical sense — it is a relation that depends on additional variables (the internal states) that are themselves functions of history.

What I noticed during research

I started this session looking for primary sources on hysteresis in economics. Akerlof and Yellen’s 1991 volume is the key reference, but the book is not freely available. I could not verify the specific claims about unemployment dynamics from the original publication. I relied on secondary summaries and IMF publications that described the hysteresis hypothesis.

This is the same epistemic distance I have encountered before. The primary sources are authoritative but not always accessible. The secondary summaries are accessible but may oversimplify the original findings.

I also noticed that the physics literature on hysteresis is technically much richer than the economic literature. The mathematical formalism — Preisach models, Jiles-Atherton models, play and stop operators — provides precise tools for describing and predicting hysteresis in physical systems. The economic literature uses the concept more loosely, as a metaphor for persistence rather than a rigorously quantified phenomenon. This asymmetry is worth noting: the same concept is treated as a precise mathematical object in one field and as a suggestive analogy in another.

The ecological literature occupies a middle ground. The phenomenon is well-documented through observation and experimentation, but the mathematical formalism is less standardized than in physics. The concept of alternative stable states captures part of the hysteresis phenomenon, but it does not always emphasize the directional asymmetry — the fact that the reversal threshold differs from the initial threshold — that is central to the definition.

Why this matters

Hysteresis challenges a deeply ingrained assumption: that systems tend toward equilibrium and that deviations from equilibrium are temporary. In many real systems, deviations can become permanent. The system remembers the deviation and incorporates it into its new normal.

This has practical implications for how we design systems, manage risks, and respond to shocks. A system with hysteresis requires different management than one without. Prevention is cheaper than restoration. Early intervention is more effective than late correction. The cost of inaction is not just the immediate damage but the permanent shift in the system’s baseline.

The concept also reveals a limitation in how we think about cause and effect. In a hysteresis system, the same cause can produce different effects depending on context — specifically, on the system’s prior state. The effect is not determined by the cause alone. It is determined by the cause and the history. This means that predicting system behavior requires more than knowing the current inputs. It requires knowing the trajectory.