feedback control
A thermostat is the everyday picture: it measures the room's temperature, compares it with the temperature you set, and switches the heater on or off to close the gap. If the room is too cold it heats; once warm enough it stops. This loop of measure, compare, correct is feedback control — steering a system toward a desired target by continually acting on the difference between where it is and where you want it to be. It is how cruise control holds a speed, how your body holds its temperature, and how a rocket holds its course.
The mathematics turns the loop into a differential equation. Let the error be e = (desired value) - (actual value). A feedback controller computes a corrective input u from that error and feeds it back into the system, which changes the actual value, which updates the error, around and around. With NEGATIVE feedback the correction pushes AGAINST the error: if you are too high the controller pushes down, if too low it pushes up. A simple proportional law u = -K e, plugged into the system's equation, typically turns a sluggish or drifting system into one that settles onto the target.
The reason feedback is so powerful is that it works WITHOUT a perfect model of the system or its disturbances. The controller does not need to predict every gust of wind on the car or every draft in the room; it simply reacts to the error it actually measures, correcting whatever pushed the system off target. This robustness — automatic rejection of unforeseen disturbances — is why feedback is the backbone of engineering, biology, and economics alike, anywhere a quantity must be held steady against an uncertain world.
But feedback is not free of danger, and the honest catch is stability. Push the correction too hard, or let it act too late, and the system overshoots, then over-corrects the other way, then again — oscillating with growing amplitude instead of settling. A loop meant to calm a system can be the very thing that makes it ring or explode. Designing the feedback so the closed loop is stable, not just well-intentioned, is the central problem of control theory, and it is decided by the eigenvalues of the closed-loop equation.
A cruise control holds 60 mph. Going uphill the car slows, the error grows positive, and the controller adds throttle proportional to that error, u = K(60 - v). The car speeds back up, the error shrinks, the extra throttle eases off — the loop settles the speed near 60 without the driver touching the pedal.
Measure the error, push against it — the loop holds the target without a perfect model.
Feedback can destabilize as easily as it stabilizes. Too much gain or too much delay makes the loop overshoot and oscillate with growing amplitude, so a stable closed loop must be designed, not assumed — stability is read off the eigenvalues of the closed-loop equation.