Feedback Loops and System Dynamics

When Outputs Become Inputs

Feedback occurs when a system's output influences its own input. It is the most powerful concept in systems thinking because it explains both stability and instability, both control and chaos. Every engineered control system uses feedback deliberately. Every runaway failure involves feedback that was not anticipated.

Negative Feedback: The Stabilizer

Negative feedback drives a system toward a setpoint. A thermostat measures room temperature (output), compares it to the desired temperature (setpoint), and adjusts heating or cooling (input) to close the gap. Cruise control measures vehicle speed and adjusts throttle. Homeostasis in biomedical systems maintains blood glucose, body temperature, and pH within tight bounds through negative feedback loops.

The defining characteristic: the system's response opposes the deviation. Temperature too high — reduce heating. Speed too fast — reduce throttle. Deviation from setpoint triggers a corrective action that reduces the deviation. The result is stability around the target.

Negative feedback is the foundation of control engineering. PID controllers (proportional-integral-derivative) are negative feedback loops with three tunable gain parameters. The proportional term responds to current error, the integral term eliminates steady-state offset, and the derivative term dampens oscillation. Nearly every mechanical and electrical control system in operation today uses some variant of this architecture.

Positive Feedback: The Amplifier

Positive feedback amplifies deviations — the system's response reinforces the change that triggered it. Compound interest: more money generates more interest, which generates more money. Viral growth: more infections produce more carriers, who produce more infections. Thermal runaway in a chemical reactor: rising temperature accelerates the exothermic reaction, which raises the temperature further.

Positive feedback is not inherently bad — it's useful when you want rapid growth or switching behavior (a flip-flop circuit uses positive feedback for bistable switching). But it is inherently unstable. Without a limiting mechanism (resource exhaustion, saturation, physical constraint, or an overriding negative feedback loop), positive feedback drives the system to an extreme — growth to capacity, temperature to destruction, voltage to saturation.

The engineering challenge is that positive feedback loops can hide in systems that were designed with only negative feedback in mind. A nuclear reactor relies on negative temperature coefficients of reactivity (temperature goes up, reactivity goes down — stabilizing). But under certain conditions, voiding of coolant can create positive feedback (less coolant means less neutron absorption means more fission means more heat means more voiding). Identifying these latent positive feedback paths is critical to safety analysis.

Delays: The Hidden Destabilizer

A feedback loop with delay can oscillate instead of stabilize — and this is where many real-world systems go wrong.

Consider a shower with a long pipe between the mixer and the showerhead. You turn the knob to hot. Nothing happens (delay). You turn it more. Still nothing. Then scalding water arrives all at once. You crank it to cold. Wait. Nothing. Then freezing water. You oscillate between extremes because the delay prevents you from observing the effect of your action in time to calibrate your response.

The same dynamic plays out in engineering systems at every scale:

• Supply chain bullwhip effect: a small demand change at the customer end creates amplifying oscillations upstream because each link reacts to delayed information with overcorrection
• Thermal lag in manufacturing processes: process control adjustments based on delayed temperature measurements overshoot and oscillate
• Software deployment feedback: a performance fix deployed today is measured next sprint — by then, new code has been added, confounding the signal

The general principle: the longer the delay relative to the feedback loop's time constant, the more likely the system is to oscillate. Short delays allow smooth convergence. Long delays produce overshoot, oscillation, or instability.

Stocks and Flows

System dynamics models systems using two primitives:

Stocks are accumulations — quantities that persist over time. Inventory in a warehouse, thermal energy in a mass, backlog of tasks, water in a reservoir. You can measure a stock at an instant in time.

Flows are rates that change stocks. Production rate fills inventory; consumption rate drains it. Heat flux raises thermal energy; cooling rate reduces it. Inflow and outflow — every stock is governed by the balance between its inflows and outflows.

Feedback loops operate through stocks and flows: a stock's level is sensed, compared to a target, and the flow is adjusted. The stock integrates the flow over time, creating the time-lag behavior that makes dynamics nontrivial. This stock-and-flow structure is why bathtub-level intuition ("is the faucet running faster than the drain?") is the right mental model for system dynamics — and why even experienced engineers get it wrong when multiple stocks interact through feedback.

Step through the stages above to compare how different feedback configurations produce fundamentally different dynamic behaviors. Notice that the same loop structure (negative feedback) produces smooth convergence or dangerous oscillation depending on whether delay is present.

Assessment