What Is a Thermostat? The Feedback Loop in Your Wall
By the BrainSnail editorial team. How these articles are written and checked, and how to tell us when one is wrong.
Set a target, measure the actual value, and act on the difference. That is the whole of feedback control, the thermostat is its most familiar example, and understanding how a simple one behaves explains why a room overshoots, why a cheap heater cycles noisily and why the fix that seems obvious, reacting faster, frequently makes things worse.
The simplest version
A classic mechanical thermostat uses a bimetallic strip, two metals with different thermal expansion bonded together, so that a temperature change makes the strip bend because one side lengthens more than the other. The bend closes or opens a contact, switching the heating on or off. This is on-off or bang-bang control, and it has one unavoidable flaw: if the switch operated at exactly the setpoint it would chatter on and off continuously as the temperature hovered there, wearing out the contacts and cycling the boiler destructively. The remedy is hysteresis, a deliberate gap between the temperature at which it switches on and the one at which it switches off, commonly created by a small magnet holding the contact until the strip pulls away sharply, or by a heat anticipator. The consequence is that the temperature never settles at the setpoint but oscillates within a band around it, and widening the band reduces wear while making the swings more noticeable.
Why rooms overshoot
Even with hysteresis, a heated room overshoots its target, and the reason is lag rather than any fault in the thermostat. Heat takes time to travel from a radiator into the air and then into the walls and furniture, so when the sensor finally reads the setpoint, a substantial amount of heat is already in transit and continues to arrive after the heating stops. The same delay works in reverse when the room cools. This combination of delay and inertia is what makes control genuinely difficult, and it explains several familiar frustrations: turning the thermostat far above the desired temperature does not heat the room faster, because the heat output is fixed and only the shutoff point changes; a thermostat placed in a draught, in sunlight, above a radiator or on an exterior wall reports a temperature the rest of the room does not share; and a system that responds very aggressively to small errors will swing further, not less, because it acts on information that is already out of date.
Better control
Industrial and modern domestic controllers go beyond simple switching, most often using proportional-integral-derivative control, whose three terms answer different questions:
- •The proportional term responds in proportion to the current error, applying more heat the further the temperature is from target, which alone leaves a persistent small offset
- •The integral term accumulates the error over time and removes that residual offset, at the cost of overshooting if the accumulated value builds up while the system is unable to respond
- •The derivative term responds to how fast the error is changing, easing off as the target is approached, which damps overshoot and amplifies sensor noise
- •Tuning the three against each other is the practical art, since a controller can be made fast, or stable, or accurate, and trading between them is unavoidable
- •Modulating heat sources, such as a boiler that varies its output rather than switching fully on and off, allow these methods to work properly, whereas a purely on-off boiler can only approximate them by varying how long it runs
What changed recently
Domestic thermostats have absorbed several ideas from elsewhere. Programmable and zoned systems heat different rooms on different schedules, which saves more energy than any change in control algorithm because the largest saving is not heating unoccupied space. Weather compensation measures outside temperature and adjusts the water temperature sent to radiators, so the system runs continuously at a low output rather than cycling hard, which suits condensing boilers that are markedly more efficient at lower return temperatures. Learning thermostats build a model of how quickly the building heats and cools and use it to start early enough to reach the target at the right time rather than when asked, which is an optimum start calculation that commercial buildings have used for decades. Occupancy sensing and phone location adjust for whether anyone is home. Heat pumps change the picture again, since they work best running steadily at low output, which makes traditional on-off thermostat behaviour actively counterproductive and is a common reason a newly installed heat pump disappoints.
The takeaway
A thermostat measures temperature, compares it to a setpoint and switches heating accordingly, which makes it the everyday example of feedback control. Simple versions use a bimetallic strip and need deliberate hysteresis to avoid chattering, so the temperature always oscillates within a band. Rooms overshoot because heat takes time to arrive and keeps arriving after shutoff, and turning the dial higher does not heat faster. Proportional-integral-derivative control, weather compensation and optimum start address the lag rather than the switching.