Balancing Emulsion Sensitivity, Grain Size, Resolution, Noise, and Reciprocity Failure

Hello. Last lesson treated the characteristic curve as a map from exposure to developed density. Two films can have curves of similar shape yet require different exposures to reach the toe, render fine detail differently, and behave differently during a long exposure. Today we will connect those measured differences to the statistics of photons and grains, the spreading of light within an emulsion, and the time needed to build a stable latent image.

The aim is to make speed, grain, resolution, and noise useful engineering quantities rather than a single “fast versus fine-grained” rule. Allow about 40 minutes, including the selected source material.

What does a faster emulsion accomplish?

A faster emulsion needs less exposure to produce a specified photographic response. On the characteristic curve, that means reaching a chosen low-density point at a smaller $H$. A speed comparison therefore requires a defined density criterion and processing conditions: changing development can shift the measured response without giving the emulsion more information from the photons that arrived.

The following demonstration, Basic Sensitometry, Part 4: Film Speed from The Naked Photographer, makes the connection to last lesson’s curve. The precise speed test is more prescribed than simply comparing two arbitrary negatives: it fixes a point near the toe and checks the curve under specified processing conditions.

For conventional black-and-white negative-film speed testing, a relevant exposure is $H_{0.1}$: the exposure producing density 0.10 above base plus fog under the prescribed conditions. In the usual lux-second convention, the speed number is proportional to its inverse:

The constant belongs to this particular speed convention; the physical point is that halving the required exposure doubles the speed number. A film rated 400 is not guaranteed to record every dark detail better than one rated 100. The rating identifies a specified response under a test procedure, while useful shadow detail also depends on contrast, fog, and noise.

Why might a grain respond at a lower exposure? From our earlier account of latent-image formation, one absorbed photon does not normally guarantee a developable crystal. Electrons and mobile silver ions must help establish a sufficiently stable cluster at a sensitivity site. Increasing a grain’s effective light-collecting area gives it more opportunities to absorb photons during the same image-plane exposure. Grain engineering and sensitization can also change how effectively those absorptions produce a developable center.

Read the archival NASA technical report’s account as a physical model, not as a claim that size alone determines speed in every modern emulsion.

A small statistical model exposes both the benefit and its cost. Suppose a grain has effective absorption area $a$, photon fluence during the exposure $F$, and absorption efficiency $\eta$. Its mean number of absorbed photons is approximately

If, as a deliberately simplified model, at least $m$ absorptions must contribute successfully to a stable center, Poisson arrival statistics give

A larger effective collecting area raises this probability at fixed fluence. But it does not add proportionally more independently located picture elements: within a fixed patch of film, larger grains generally mean fewer possible grain positions. Real emulsions also contain distributions of grain sizes and sensitivities, and the probability that an absorption survives as a latent image is not fixed. The model is a way to isolate the trade-off, not a formula for predicting an ISO rating.

The record is made of discrete, variable grains

A developed negative is not a continuous deposit of darkness. At a given exposure, some crystals develop and others do not; the resulting silver is distributed in discrete, irregular structures. Viewed at a scale large compared with those structures, a patch has a mean density. Viewed more closely—or enlarged substantially—its density fluctuates. That fluctuation is granularity; its visibility in a final picture is commonly called graininess.

There are several linked sources of variation. Photon arrivals fluctuate even under perfectly steady illumination. Crystals differ in size, sensitivity sites, and location. Development turns successful latent images into visible material, but its amplification does not reconstruct the precise positions of the photons responsible. Unexposed grains that develop contribute fog. These effects do not all enter as separate noise terms that can automatically be added: for example, a grain’s probability of developing already includes photon-arrival randomness.

Consider $M$ otherwise identical, independently responding grains in a small image area. If each develops with probability $p$, the expected developed count and its variance are

This toy calculation shows why the number of independent opportunities matters. At comparable fractional response, fewer, larger grain sites produce larger relative count fluctuations. In an actual film, grains are unequal, development and optical interactions introduce correlations, and transmitted-light density is not simply proportional to $K$. Nevertheless, coarse granularity is one common price of obtaining a response from fewer incident photons.

Noise must also be judged against the signal of interest. For two nearby exposures, useful tonal separation is the difference in their mean densities—or, in the model, the difference in expected grain counts—relative to fluctuations within each patch. Near the toe, a small density change may be especially hard to distinguish from grain-to-grain variation and fog. This is why merely making a low-exposure patch develop is not the same as recording fine shadow texture reliably.

Historical measurements report the familiar tendency of very fast materials to have more prominent grain and poorer resolution, but also warn that it is not an exceptionless rule.

One caution follows directly from the previous lesson: greater development may increase density or contrast and make grain more conspicuous. It cannot restore photons that were never captured, or shadow distinctions absent from the latent image. Thus effective speed after a processing change and information recorded at exposure should not be conflated.

Why grain size does not set resolution by itself

Resolving power concerns whether a film can preserve distinct fine features—for example, alternating dark and light lines. It is often quoted in line pairs per millimetre, but the result depends on the test pattern’s contrast and on the criterion for calling the lines distinguishable. At 100 line pairs per millimetre, a full dark-plus-light cycle occupies $10\mu \mathrm{m}$ on the film. That number does not mean every grain must be $10\mu \mathrm{m}$ wide: many grains may participate in recording each cycle.

The image can lose fine contrast before grain size becomes a simple geometric limit. Light may scatter or reflect as it traverses the turbid emulsion, exposing grains adjacent to where the lens first delivered it. Development can spread the recorded edge further. Increasing exposure can make this spreading more consequential. Meanwhile, the random arrangement of developed grains makes a weak, high-frequency pattern progressively harder to distinguish from granularity.

The NASA report distinguishes limiting resolution from acutance, the apparent abruptness of an edge. Read its physical explanation of scattering alongside those definitions.

The iodine depth-series image below illustrates a further reason not to model every grain as an identical sphere. Its changing interior contrast and conspicuous grain boundaries show structure that varies with position and depth. It is an ion-microprobe view of silver bromoiodide grains, not a direct picture of developed silver or a measurement of their photographic speed.

A useful modern description is to ask how much contrast the processed film transfers at each spatial frequency, and how much granularity competes with that transferred contrast. A single limiting line-pair number cannot answer both questions. High edge contrast may look sharp—high acutance—without preserving every subtle fine texture. Conversely, a fine-grained material may have ample potential resolution but still fail to show low-contrast detail when too few photons have exposed it. The lens and the emulsion both contribute to the final image; the film’s quoted resolution is not automatically the camera’s resolution.

Equal total exposure need not mean equal photographic effect

So far, exposure has been written $H=Et$: illuminance multiplied by time. The reciprocity law would say that, for a given emulsion, any two combinations with the same $H$ produce the same response. For instance, a weaker light used for proportionally longer should leave the characteristic curve unchanged. Film obeys this approximately over a useful range of exposure times, but not indefinitely.

The reason is that photons must help establish a stable latent-image center, not merely arrive at some point during the exposure. At very low arrival rates, an incipient silver cluster can disperse before further successful events enlarge it. The same total photons delivered faster may form more developable centers. The result is low-intensity, long-exposure reciprocity failure: a metered exposure time calculated from $H=Et$ yields less density than expected. Increasing time beyond the meter’s recommendation can compensate, but the correction depends on the material and exposure regime.

Very brief, intense exposures can fail for a different reason. In the account below, rapid generation of electrons favors numerous small centers rather than enough silver accumulating at individual centers to make them developable. This high-intensity, short-exposure failure is not the same mechanism as the long-exposure loss, and neither should be assumed equally important for every film.

This is a particularly consequential limitation for long exposures of faint scenes. The meter estimates incident exposure, whereas the emulsion’s usable response depends on when absorptions occur. A manufacturer’s measured compensation data are preferable to extrapolating from an ISO number: ISO speed alone does not specify the long-exposure correction. Processing can change the resulting densities, but it cannot make the reciprocity assumption physically true.

Conclusion

Emulsion sensitivity is the exposure required to reach a defined processed response. Larger effective grain collecting areas can help a film respond to fewer photons, but typically reduce the number of independently located recording sites in a small area. That can increase visible granularity; it does not, by itself, fix resolving power. Scattering, development, target contrast, exposure, and the signal-to-noise ratio of fine detail also matter.

Finally, $H=Et$ describes the incident light, not a guarantee of equal latent-image formation at every intensity and time. Long, dim and extremely short, intense exposures can depart from reciprocity for different physical reasons. Next we will move from emulsion performance to the materials and manufacturing choices that made different photographic formats practical, from individual plates to flexible roll film.