The Exposure Triangle Is Three Routes to the Same Stop

A stop is a factor of two, and that is the whole unit

Photography measures light in stops, and a stop means exactly one thing: twice as much light, or half as much. It is a ratio, not an amount, which is why the same word works at midday and at midnight. Two of the three controls implement it in the obvious way. Doubling the exposure time from 1/256 s to 1/128 s doubles the light that reaches the sensor, because the sensor is simply collecting for twice as long. Doubling the ISO setting from 100 to 200 doubles the brightness of the resulting file, because it doubles the gain applied to whatever the sensor collected.

The aperture does not, and that single fact is the source of every confused explanation of the f-number scale. The aperture is a hole. The light passing through a hole is proportional to its area, and area goes with the square of the diameter. But the number engraved on the barrel is not an area and not a diameter — it is N = f/D, the focal length divided by the diameter of the entrance pupil. It is a ratio, it is dimensionless, and it is inverted: bigger number, smaller hole.

Deriving the f-number scale instead of memorising it

Start from the geometry. The pupil diameter is D = f/N, so the pupil area is π(f/2N)², which is proportional to 1/N². The light gathered therefore scales with 1/N². Now ask what N does when you want half the light: you need 1/N² to halve, so N² must double, so N must be multiplied by √2. That is the entire derivation. The sequence of full stops is N = (√2)^k = 2^(k/2), starting from N = 1.

Run k from 0 upward and the familiar numbers fall out: 1.00000, 1.41421, 2.00000, 2.82843, 4.00000, 5.65685, 8.00000, 11.31371, 16.00000, 22.62742. The even-k terms are exact powers of two and are marked exactly. The odd-k terms are irrational and are rounded down for legibility. f/1.4, f/2.8 and f/5.6 are each 1.01% below their true value — 0.029 of a stop, which no film or sensor has ever noticed. f/11 and f/22 are rounded harder, 2.77% low, or 0.081 of a stop.

That rounding has a visible consequence on the barrel. The step from f/8 to f/11 is only 0.919 of a stop, and the step from f/11 to f/16 is 1.081 — the deficit and the surplus cancel, so f/8 to f/16 is exactly two stops as it must be. The same holds for f/16, f/22, f/32. The errors never accumulate because the marked value is always compared back to the exact geometric term, not chained from the previous marking.

Exposure value puts one number on the two mechanical axes

Aperture and shutter are the two settings that actually change how much light lands on the sensor, and exposure value combines them: EV = log2(N²/t), with t in seconds. Because it is a base-2 logarithm, one unit of EV is one stop, and the same number can be reached from either axis. Reading the table above: f/4 at 1/256 s is EV 12, and so is f/8 at 1/64 s, and so is f/22 at 1/8 s. All three deliver the same quantity of light.

Two checks confirm the formula rather than just asserting it. Sunny 16 — the field rule that on a clear day you can shoot at f/16 with the shutter set to the reciprocal of the ISO — gives f/16 at 1/128 s at ISO 100, and log2(16²·128) = log2(32768) = 15 exactly. And EV converts to actual scene luminance through the meter calibration constant: L = K·2^EV/S with K = 12.5 and S = 100 gives 4096 cd/m² at EV 15, which is the published figure for that light level. The formula is not a convention layered on a convention; it lands on measurable candelas.

One honest footnote about the shutter dial. The exact ladder at EV 12 is 1/4096, 1/2048, 1/1024, 1/512, 1/256, 1/128, 1/64, 1/32, 1/16, 1/8 second — every value a clean power of two. Your camera prints 1/4000, 1/2000, 1/1000, 1/500, 1/250, 1/125, 1/60, 1/30, 1/15, 1/8. Those are the same rounding convention applied to the time axis that the barrel applies to the aperture axis. The marked 1/250 is really 1/256 of a second, and 1/60 is really 1/64. Neither the maths nor the picture cares; but if you compute EV from the printed numbers you will be a few hundredths of a stop out, and now you know why.

One scene, three ways, three different photographs

Take a real scene: a person walking, framed with a 50 mm lens on full frame at 10 ft, metered at EV 12, exposed at f/4 and 1/256 s and ISO 100. Now a cloud arrives and the light drops one stop, to EV 11. You have three ways to give the sensor that stop back, and the exposure triangle says they are interchangeable. They are, in brightness. In every other respect they produce three different photographs.

Route one, aperture: open from f/4 to f/2.8. EV becomes log2(2.828²·256) = 11, correct. But depth of field at 10 ft falls from 33.9 in to 23.5 in, a 31% loss. If the subject is at an angle, or if two people are standing a foot apart, one of them just went soft.

Route two, shutter: slow from 1/256 s to 1/128 s. A person walking at 3 mph, crossing the frame at 10 ft with a 50 mm lens, is magnified 1:60 onto the sensor, so their image moves 87 µm during 1/256 s and 175 µm during 1/128 s. Against a circle of confusion of 28.8 µm, that is 3.0 times the sharpness criterion before and 6.1 times after — 14.6 pixels then 29.1 pixels on a 24-megapixel full-frame sensor with a 6.0 µm pitch. The subject was already smeared; you just doubled the smear.

Route three, ISO: raise from 100 to 200. Nothing optical changes, and that is precisely the problem — the sensor still receives exactly the EV 11 amount of light, half what it got before, and the gain merely scales the result back up to the same brightness. Photon shot noise scales with the square root of the photon count, so a patch that collected 10,000 electrons at ISO 100 collects 5,000 at ISO 200: signal-to-noise falls from 100:1 to 70.7:1, from 40.00 dB to 36.99 dB. Every stop of ISO costs 3.01 dB, permanently, in the shadows first.

ISO is not sensitivity, and calling it that hides what it does

On film, sensitivity was physical: a faster emulsion had larger silver halide crystals that needed fewer photons to develop, and the grain you saw was those crystals. A digital sensor has no such variable. Its quantum efficiency — the fraction of arriving photons that become measurable electrons — is fixed by the silicon and does not move when you turn the dial. Turning the ISO dial to 200 does not make a photosite catch more light; it makes the camera amplify what the photosite caught.

That amplification is not useless. Applied in the analogue domain, before the analogue-to-digital converter, it lifts the signal above the fixed read noise of the downstream electronics, which is why raising ISO in camera can genuinely beat shooting dark and lifting the file afterwards. On sensors described as ISO-invariant, that downstream noise is already so low that the two routes converge and the choice barely matters. What no amount of gain can do is invent photons: the shot noise was set the moment the shutter closed, and every stop of ISO you spend is a stop of light you decided not to collect.

Choosing which axis to spend

Because the three are not interchangeable in the picture, the decision is never about exposure — it is about which artefact you are willing to accept. Start from the constraint that is non-negotiable in the shot. If the subject moves, the shutter is fixed first and the other two absorb the difference. If two people must both be sharp, the aperture is fixed first. Only what is left over goes to ISO, because ISO is the axis with no upside at all: it buys nothing except brightness, and it pays in noise.

The fourth option the triangle never mentions is to add light or change position. Moving from 10 ft to 20 ft, as the depth-of-field article shows, multiplies the sharp zone by 4.29 without spending a single stop. Bringing the subject nearer a window does the same for brightness. The triangle is a good description of a closed system with three dials; the photograph is not a closed system.

Frequently asked questions

Is f/11 really exactly one stop from f/8?

Not as marked. The exact term is 8·√2 = 11.31371, so the engraved 11 is 2.77% low, which is 0.081 of a stop. The step from f/8 to f/11 is therefore 0.919 stops, and the step from f/11 to f/16 is 1.081 stops. They cancel: f/8 to f/16 is exactly two stops. Inside the lens, the mechanism is built to the exact geometric value, not to the printed one, so the aperture the camera sets is right even though the label is rounded.

Why does my camera say 1/250 when the maths wants 1/256?

Same rounding convention as the aperture ring, applied to time. The exact halving ladder is 1/8, 1/16, 1/32, 1/64, 1/128, 1/256, 1/512, 1/1024, 1/2048, 1/4096 second. Cameras print 1/8, 1/15, 1/30, 1/60, 1/125, 1/250, 1/500, 1/1000, 1/2000, 1/4000 because those are shorter to read. The mechanism targets the power of two. If you compute EV from the printed numbers you will be off by a few hundredths of a stop, which is invisible; if you are debugging a meter or writing a calculator, use the powers of two.

If ISO is only gain, should I shoot at base ISO and brighten the file afterwards?

Sometimes, and it depends on where the noise floor of your camera sits. Gain applied before the analogue-to-digital converter lifts the signal above the fixed read noise of the electronics that follow, so on cameras with significant downstream noise, raising ISO in the field beats lifting a dark raw file later. On sensors described as ISO-invariant, that downstream noise is small enough that the two approaches converge and you may as well stay at base ISO and keep the highlight headroom. Neither route recovers photon shot noise, which was fixed when the shutter closed.

Does exposure value already include ISO?

No. EV = log2(N²/t) contains only the aperture and the shutter, which is exactly right, because those are the two settings that change how much light reaches the sensor. ISO changes how that light is turned into a file. By convention EV figures are quoted at ISO 100, written EV₁₀₀, and moving to another speed shifts the whole mapping by log2(S/100) stops. The scene luminance behind an EV number comes out of L = K·2^EV/S, with the reflected-light calibration constant K set at 12.5 by most manufacturers: EV 15 at ISO 100 is 4096 cd/m², EV 3 is 1 cd/m².

So is the exposure triangle wrong?

It is incomplete rather than wrong. As a bookkeeping device for brightness it is exact: any stop taken from one control can be handed to another and the file comes out the same brightness, and the calculator does that arithmetic reliably. What it omits is that each control has a second, unrelated output — depth of field, motion blur, noise — and that those second outputs are the reason you chose the setting in the first place. Treat the triangle as the constraint you must satisfy, and the three side effects as the thing you are actually deciding.