Reciprocity failure
A Socratic walk-through of reciprocity failure — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why does doubling an already long film exposure fail to double what the negative records?
Photography rests on a bargain so tidy it feels like arithmetic: halve the light, double the time, and the negative comes out the same. Photographers work it constantly.
Then the exposures get long. A moonlit landscape metered at thirty seconds comes out thin; give it a minute and it is still thin; give it four and it may finally be right. Somewhere between a five-hundredth of a second and half an hour the arithmetic quietly stopped applying — on film, while leaving a digital sensor on the same scene unbothered. What can film tell between a lot of light quickly and a little light slowly, when the total is the same?
Reasoning it through
REASONING #The bargain — that only the product of illuminance and time matters — is the reciprocity law, and it would be exactly right if the emulsion were a bucket collecting photons. So: is it a bucket?
It is not. A film emulsion is millions of silver halide microcrystals in gelatin, and each grain does something more like making a decision than filling up. A photon absorbed by a grain frees an electron; the electron is trapped at a crystal defect, where it neutralises a mobile silver ion, leaving one neutral silver atom. That atom begins a latent image speck — and a lone atom is unstable, reverting shortly so the grain is back where it started.
Only when several such atoms accumulate at the same site does the speck become stable. The classical Gurney-Mott account puts the minimum at a small handful — three or four is the usual figure, which I give as recalled rather than derived. Once a grain has that speck, development converts the entire grain to silver; without it, the grain develops not at all. So density is not a measure of photons absorbed but a count of grains that crossed a threshold — and a threshold partial progress can leak away from behaves quite differently from a bucket.
Now run the two regimes. In bright light, photons arrive at a grain in rapid succession, so the second atom forms before the first has reverted and the third before the second: the speck builds and the grain latches. In very dim light, the interval between hits exceeds the lifetime of the sub-critical speck, so the grain accumulates an atom, loses it, accumulates another, loses it. It can absorb a great deal of light and never latch — inefficient not because it received less light but because it came too slowly to hold on to.
Photographers describe this with the Schwarzschild relation: effective exposure goes as illuminance times time raised to a power slightly less than one, the exponent usually quoted around 0.7 to 0.9 — a recalled range, genuinely film-specific. Take 0.8. Doubling the time multiplies the effect by two to the power 0.8, about 1.74 rather than 2; to actually double it you need time multiplied by two to the power of one over 0.8, about 2.4 — the metered thirty seconds becoming well over a minute. Quadrupling needs about 5.7 times.
There is a second failure at the opposite end. At extremely high intensities — a very brief, very bright flash — efficiency also drops, because so many electrons are liberated at once that they form several scattered specks rather than one reaching critical size. The law holds best in the middle, where ordinary photography lives.
The analogy
THE ANALOGY #Think of starting a snowball by hand. Pack snow on fast enough and it consolidates into a core that holds together and takes everything you add afterwards. Pack it too slowly, in a thaw, and each handful softens before the next arrives — you can work all afternoon, move an enormous amount of snow, and have no snowball. What matters is not how much snow you handled but whether you got past the size at which it holds.
A snowball grows continuously and its critical size is soft and fuzzy, whereas a grain's decision is all-or-nothing — it develops fully or not at all — and a photograph's smooth greys are millions of those binary verdicts averaged by the eye, not any grain partly exposed.
Clarifying the model
THE MODEL #First, the misconception. Reciprocity failure is not the film "getting tired"; nothing is used up and no fatigue accumulates. The emulsion is as sensitive as ever. What changed is the arrival rate, and a threshold that decays is rate-sensitive by construction.
Second, a consequence that confirms the mechanism from an unexpected direction. Colour film carries three emulsion layers whose exponents are not identical, so a long exposure does not merely underexpose — it underexposes the layers by different amounts and the image shifts in colour. That the failure produces a colour cast and not only density loss is hard to explain on any account not located in each emulsion's chemistry.
Third, the digital contrast. A silicon photodiode integrates charge linearly: an electron freed is an electron stored, with no unstable intermediate to decay. There is no reciprocity failure in the film sense, and the limit on a long digital exposure is dark current and read noise — additive contamination, not a breakdown of the law.
Fourth, a boundary against a neighbour using the same word. Museum lighting rests on a lux-hours dose model whose working assumption is precisely that reciprocity holds: halving illuminance and doubling exposure does the same damage. That is a conservation convention, roughly right, known to break down for some materials. The fixed point of difference: there, reciprocity is an assumed damage model for an object being slowly destroyed; here, a measured response law of an object being recorded on. Same arithmetic, opposite intent.
How would we know? Expose a sensitometric wedge at matched total exposures across a wide range of intensities — a thousandth of a second at high illuminance, a hundred seconds at very low — and compare the density curves. The reciprocity law predicts they superimpose; this account predicts the long dim exposures fall progressively short, and a silicon sensor does not. The refuting observation: film curves superimposing across intensities, or a linear sensor showing the same shortfall.
A picture of it
THE PICTURE #How to readThis is a journey chart repurposed — the score is not satisfaction but how faithfully the reciprocity law holds, 5 meaning doubling the time doubles the record exactly and 1 meaning it barely helps. Read left to right as the exposure gets longer and dimmer. The film line starts poor at the brief bright end, is near-perfect through the middle where ordinary photography sits, and falls away steeply into long dim exposures — a mechanism failing at both extremes for different reasons. The sensor line stays flat because it has no unstable intermediate to lose; its long-exposure trouble is noise, not a failure of the law.
What became clearer
WHAT CLEARED #Film does not collect light; it counts grains that got over a wall. A grain climbs by accumulating silver atoms at a site, and a partial climb slides back if the next photon takes too long. So the emulsion is sensitive not just to how much light it receives but to the rate — and in dim light, where the intervals outlast the memory of the partial speck, an ever-larger share of the light is spent on progress the grain cannot keep. The tidy bargain was never a law; it was the good behaviour of a threshold process in the middle of its range.
Where to go next
ONWARD #- Why the exponent differs between emulsions, and what a manufacturer changes to move it.
- Why correcting a long colour exposure needs a filter as well as extra time.
Key terms
TERMS #| Term | What it means |
|---|---|
| Reciprocity law | the assumption that photographic effect depends only on illuminance times time. |
| Latent image speck | the cluster of silver atoms on a grain that, once stable, makes the whole grain developable. |
| Schwarzschild exponent | the power applied to time in the corrected exposure relation, below one for emulsions showing low-intensity failure. |
| Dark current | charge accumulating in a sensor without light; the limit on long digital exposures. |
Every term the collection defines is gathered in the glossary.