Opening the lid on the magic
Back in the ideal rung you watched something remarkable: the gain of every op-amp circuit came out as a plain ratio of two resistors, while the chip's own enormous gain vanished from the answer. We called that a revolution and waved at the mechanism — negative feedback, the tireless thermostat — without prying it open. This guide pries it open. And inside sits a single number, the loop gain, that quietly decides how precise, how linear, how wide-band, and how stiff your amplifier turns out to be.
Meet the cast. The op-amp's raw, datasheet gain — with no feedback wrapped around it — is the open-loop gain A: huge (often 100,000 to over a million), but sloppy. It varies wildly from chip to chip, drifts with temperature, sags as the signal swings, and collapses as frequency rises. The gain you actually keep, set by your own feedback network, is the closed-loop gain G: small, but rock-solid. The feedback network samples a fraction of the output — call that fraction b — and subtracts it from the input. The product A times b is the loop gain: the gain a signal sees on one trip all the way around the loop.
The one equation behind it all
Let us actually write it down — it is one line of algebra, and it explains everything that follows. Feed your signal Vin in, let the fraction b of the output return and be subtracted, so the op-amp sees the error Vin - b times Vout and multiplies it by A: Vout = A times (Vin - b times Vout). Gather the Vout terms onto one side and you have the master equation: G = Vout / Vin = A / (1 + A times b).
Now stare at the denominator. The term A times b is the loop gain, T. When the loop gain is large — and with A near 200,000 it is enormous — the 1 becomes a rounding error and G collapses to simply 1/b. The op-amp's own unreliable A has cancelled itself out of the answer. That is the whole trick in one move: a circuit whose behaviour would depend on a sloppy, drifting A is rebuilt into one whose behaviour depends only on b, which you set with two precise resistors. The closed-loop gain is whatever you chose 1/b to be — no more, no less.
Put real numbers on it. Say you want a gain of 10, so 1/b = 10 and b = 0.1 (Rin = 10 k, Rf = 90 k). With A = 200,000 the loop gain is T = A times b = 20,000, and G = 200,000 / (1 + 20,000) = 9.9995 — short of the ideal 10 by only 0.005%. Now grab a worse chip whose A is half as good, or let it droop with heat: even at A = 400,000 (double) G is 9.99975, and at A = 100,000 (half) it is 9.999. A factor-of-four swing in the raw part moves your gain in the fourth decimal place. That is the immunity feedback hands you, and the worked example you should keep in your head.
Loop gain is a budget — here is what it buys
Everything good about feedback scales with that one number, the loop gain, so it pays to picture it as a budget you spend. First, precision. The fractional gain error — how far your real gain sits below the ideal 1/b — is almost exactly 1 / (1 + T). With T = 20,000 that error is 0.005%, and, better still, it stays tiny even as A misbehaves: double A and the closed-loop gain shifts by a few parts per million. Feedback divides the op-amp's sloppiness by (1 + T). That is why you can build a gain of exactly 10 from a part whose own gain you do not trust to even one significant figure.
The same divisor cleans up more than gain. Any distortion the op-amp generates inside the loop — the kinks and curvature of its own transfer characteristic — is shrunk by (1 + T) too, because the loop sees the distorted output, compares it against the clean input, and corrects the difference. So is the output impedance: a bare op-amp output of, say, 75 ohm becomes 75 / (1 + T), only a few milliohms, a far stiffer source. Push more loop gain at the problem and precision, linearity, and stiffness all improve together. They are not three separate features — they are the same surplus of gain, spent three ways.
Watching loop gain on a Bode plot
That phrase 'at the frequencies that matter' is the catch, because loop gain is not a constant — it melts away as frequency climbs. Draw it on a Bode plot, gain in decibels against log frequency. The open-loop gain A starts huge at DC and rolls off in a straight 20 dB-per-decade slide. Your flat closed-loop line sits across it at 1/b. And here is the lovely part: the vertical gap between the two curves IS the loop gain in dB. It is fat at DC and narrows as the falling open-loop curve descends toward your flat line.
gain
(dB)
106 |\ A = open-loop gain
| \ (DC ~106 dB, falls 20 dB/decade)
| \ ^
| \ | loop gain T (the GAP between the curves)
20 |----\-v------------------- 1/b = closed-loop gain (20 dB = x10)
| \ flat, the gain you keep
| \
0 |.......\.................. T = 0 dB: gap closed, feedback gone
+--------------------------> frequency (log scale)
5 Hz 100 kHz 1 MHz
the gap = loop gain: ~86 dB at DC, zero where they cross at 100 kHz.Where the two curves meet, the gap has closed to zero: loop gain is 1, and past that point feedback has nothing left to spend. That crossover sets your closed-loop bandwidth, and it lands at a tidy place — the gain-bandwidth product divided by your closed-loop gain. A 1 MHz part set for a gain of 10 stays flat only to about 100 kHz; ask for a gain of 100 and you get just 10 kHz. Gain and bandwidth trade one for one, because both are drawn from the same shrinking pool of loop gain. The next guide makes a whole study of these AC limits.
The honest fine print
Three honest qualifications keep this from sliding into folklore. First, feedback spends gain — it never creates it. You begin with a vast open-loop A and trade most of it away for precision and bandwidth; you cannot end up with more total gain, or more output power, than the raw chip could deliver. Second, the golden rules from the last rung — V+ equals V-, the virtual short — are the idealized limit of infinite loop gain. With finite loop gain a real residual difference survives: V+ - V- = Vout / A, a few microvolts at DC, but proportionally larger as A falls with frequency. The rules are an excellent approximation, not a law of nature.
Third, and most important: you cannot simply pile on loop gain without limit. Every extra stage and every parasitic capacitor adds phase shift around the loop, and once the returning signal is delayed by half a cycle, your negative feedback arrives as positive feedback — the thermostat now shoves in the wrong direction, and the amplifier oscillates. So loop gain is capped not by greed but by stability, the subject of guide 4. Two final honesties: feedback cannot outrun the slew rate, a large-signal limit on how fast the output can move at all; and it reduces distortion without abolishing it — and because loop gain shrinks with frequency, a tone at 10 kHz comes out dirtier than the same tone at 100 Hz.