A difference, amplified absurdly hard
An operational amplifier (op-amp) is, at heart, one job done to an extreme: it looks at the difference between its two inputs and multiplies that tiny difference by an enormous number. The two inputs are the non-inverting input (marked +) and the inverting input (marked -), and the rule is simply Vout = A times (V+ - V-). You already met the seed of this in the last rung — the differential pair of two matched transistors that subtracts one input from the other. An op-amp is that idea finished off and polished: a whole multi-stage amplifier baked into one chip.
How big is that multiplier A, the open-loop gain? Absurdly big — often 100,000 to over a million. That sounds wonderful until you do the arithmetic. With A = 200,000 and a 15 V supply, the output hits its limit (the rail) once the input difference reaches just 15 / 200,000 = 75 microvolts. A breath of difference — a stray 0.1 mV of noise — pins the output hard against one rail. Raw, an op-amp is a hopelessly twitchy thing: you cannot use it on its own to make a clean amplifier of gain 10.
The two golden rules
To tame that twitchiness we first idealize the part. The ideal op-amp has three perfect properties: infinite open-loop gain, infinite input resistance (so the inputs draw no current at all), and zero output resistance (so the output is a perfect voltage source, unbothered by its load). Real chips only approximate these, but the approximation is so good that designing with the ideal model first, then checking the small corrections later, is exactly how professionals work.
From that ideal model fall the two golden rules — the two sentences that let you analyze almost every op-amp circuit in your head. Rule 1: no current flows into either input. (Infinite input resistance means the inputs are perfect spectators.) Rule 2: the output does whatever it must to drive the two inputs to the same voltage, V+ = V-. This second one feels like magic, so picture a negative-feedback thermostat: it watches the room and forever nudges the heater so the actual temperature matches the dial. The op-amp watches its own two inputs and forever nudges its output so they match.
Three circuits the rules build instantly
Watch the rules do all the work on the most important circuit, the inverting amplifier. Tie the + input to ground (0 V), feed the signal Vin through an input resistor Rin into the - input, and bridge a feedback resistor Rf from the output back to that same - input. The next guide is devoted to this circuit; here, walk the derivation once so you feel how mechanical it is.
- Find the - input's voltage. The + input is at 0 V, so by Rule 2 the output drives the - input to 0 V as well — even though no wire connects it to ground. This is the famous virtual ground: a node held at 0 V purely by the op-amp's effort, not by a real connection.
- Find the current through Rin. It flows from Vin to the virtual ground (0 V), so by Ohm's law it is I = (Vin - 0) / Rin = Vin / Rin.
- Route that current into Rf. By Rule 1 none of it leaks into the - input, so every bit must continue on through Rf to the output. The same I flows through both resistors in series.
- Read off the output. Going from the virtual ground (0 V) across Rf, the output sits at Vout = 0 - I times Rf = -(Vin / Rin) times Rf. So the gain is Vout / Vin = -Rf / Rin. With Rin = 10 k and Rf = 100 k, the gain is -100/10 = -10: feed in 0.2 V and get out -2.0 V. The minus sign just means the output is upside-down relative to the input.
Move the signal to the + input instead and you get the non-inverting amplifier, whose gain is 1 + Rf/Rin and is always positive (output the right way up). With Rf = 90 k and Rin = 10 k the gain is 1 + 90/10 = 10. Shrink Rf to zero (a plain wire) and Rin to infinity (remove it) and the gain becomes exactly 1: that is the unity-gain buffer, whose output simply copies its input. A gain of 1 sounds useless until you remember the op-amp's input draws no current and its output is stiff — so a buffer is a perfect go-between, like choosing a hose that neither chokes a trickling source nor starves a thirsty load. It is the classic fix for the loading problems you met back in the DC rungs.
More building blocks from the same two rules
The virtual ground is a gift that keeps giving. Bring several signals, each through its own resistor, into that single virtual-ground node and you have a summing amplifier. Because the node sits steadfastly at 0 V, each input pushes its own current Vk/Rk completely independently of the others; Rule 1 forces all of them to merge into Rf, so Vout = -(Rf/R1 times V1 + Rf/R2 times V2 + ...). With Rf = R1 = R2 = 10 k and V1 = 1 V, V2 = 0.5 V, the output is -(1 + 0.5) = -1.5 V. That is literally an audio mixer: many sources added with their own volume knobs, and not one of them feels the others.
Put one signal on each input through a matched set of four resistors and you get the difference amplifier (the subtractor): its output is proportional to V+ - V-, and any voltage common to both inputs cancels out. That last trait is precious — it rejects the hum and noise that ride equally on a pair of sensor wires. Stack a few op-amps to do the same job with huge, easily-set gain and near-perfect rejection and you have the instrumentation amplifier, the front end of nearly every precise measurement.
Now swap a feedback resistor for a capacitor and the circuit starts doing calculus. Replace Rf with a capacitor across the inverting input and the steady input current charges it at a constant rate: the integrator outputs Vout = -(1/RC) times the running integral of Vin — a perfect ramp for a constant input, like the virtual ground patiently filling a bucket. Put the capacitor at the input instead and you get the differentiator, whose output tracks how fast the input is changing, -RC times dVin/dt. Guide 5 of this rung returns to both in earnest.
CIRCUIT where the signal enters output Vout ------------------ ---------------------------- -------------------------- inverting amp Vin -> Rin -> (-) node -(Rf/Rin) times Vin non-inverting amp Vin -> (+) input (1 + Rf/Rin) times Vin unity buffer Vin -> (+), out tied to (-) +1 times Vin (a copy) summing amp V1,V2 -> R1,R2 -> (-) node -(Rf/R1 V1 + Rf/R2 V2) difference amp V- on (-), V+ on (+), 4 R's (V+ - V-) matched R's integrator Rin -> (-), C as feedback -(1/RC) integral of Vin dt differentiator C -> (-), Rf as feedback -RC times dVin/dt In EVERY row: (+) and (-) are held equal, and no current enters either input.
The revolution, and the honest fine print
Stand back and notice what just happened. In every one of those circuits, the gain is set by a ratio of two resistors — and the op-amp's own gargantuan, sloppy open-loop gain dropped out of the answer entirely. That is the revolution. Last rung you bent over backwards to stop a transistor's wildly-varying beta from setting the gain; here, negative feedback trades away the op-amp's huge-but-unreliable gain in exchange for a small, rock-solid gain you dial in with two cheap, precise resistors. Feedback converts raw power into precision — that is the deepest idea in all of analog electronics.
But be honest about what the golden rules quietly assume. They hold only with negative feedback, and only inside the chip's real limits. The output cannot swing past the supply rails. It cannot change faster than the slew rate (a 0.5 V/us part simply cannot make a clean 5 V step in less than 10 us). And gain is borrowed against bandwidth through the gain-bandwidth product: a 1 MHz part set for a gain of 10 runs out of gain at just 100 kHz. Push too much phase shift around the feedback loop and the thermostat overshoots, then overshoots its overshoot — the amplifier oscillates instead of settling. Keeping the loop calm is the whole subject of amplifier stability, which the 'real op-amp' rung tackles next.