Optical sizes in type

Why does a typeface that reads beautifully at footnote size look clumsy blown up on a poster?

Letters are shapes, and a shape enlarged is the same shape. Scale a footnote up to poster height and every proportion inside it survives exactly: thick to thin, the size of the counters, the gaps between the letters. Nothing has been distorted.

And yet the enlarged footnote looks wrong — lumpen, gappy, faintly agricultural — while a face drawn for posters, shrunk to footnote size, turns into grey mud. If the geometry is untouched in both directions, the fault cannot be in the geometry. So what changes when the size changes?

Ask the question the other way round. What in the reading situation fails to scale when the letters do?

Two things, and they pull the same way. The first is the machinery that puts ink or pixels on the surface. Ink spreads into paper fibres by a small absolute amount at every edge, and every imaging process has a comparable fixed absolute error. That error does not shrink with the type. Take an illustrative figure — twenty micrometres of spread at each edge, the right order for ordinary printing though it varies enormously with paper and press. A hairline in six-point type might be a twentieth of a millimetre wide; twenty micrometres from each side nearly doubles it, or chokes a counter shut. Scale that letter twelve times to seventy-two point and the hairline is over half a millimetre, where the same twenty micrometres is a few percent and invisible.

The second is the eye, likewise fixed. Acuity and crowding — the way neighbouring letters interfere with identification — are limits in angular terms, and a reader sits the same distance from a caption as from a headline. At six point the letters are close to those limits, so the design must spend everything on separability: more space between letters, a larger lowercase relative to the body, open apertures so a c does not read as an o, sturdier thin strokes so they survive at all. At seventy-two point the reader has resolution to spare, and can be given delicacy instead.

Put the two together and you have the answer. A caption face is not a display face that happens to be small; it is a set of compensations for problems that only exist when small. Enlarge it and the compensations enlarge too, but the problems they corrected have gone. What is left is a correction with nothing to correct: thick thins that now look coarse, blunt heavy serifs, spacing that reads as holes. The clumsiness is a solution still visible after its problem has been removed.

Run the reasoning backwards and the other failure explains itself: a shrunk display face has hairlines below what the ink or the eye can hold, so they break or fill in, and its tight fit crosses the crowding limit.

Is any of this visible in the historical record? Neatly, because metal type could not have done it another way. Every size had to be cut as a separate set of punches by hand, so a six-point and a thirty-six-point of the "same" type were always different drawings — and founders used that necessity, making small sizes sturdier and wider-fitted and large ones finer. One outline fitting all sizes arrived only with photographic and then digital setting, and it is precisely then that the problem appears as a defect rather than a fact of the craft. The correction has since been rebuilt deliberately, as separate caption, text and display cuts, and as the optical-size axis of a variable font.

What would show this wrong? Take one family with a genuine caption cut and a genuine display cut, set the same passage in each at each size, and ask readers to choose. The account predicts the preference must reverse with size; if readers simply prefer whichever cut they have seen more often, the story is familiarity rather than optics. A second discriminating test: if the whole effect were ink spread, a high-resolution screen with no spread should abolish the need for optical sizes. It does not — small type on a good screen still wants the sturdier, wider-fitted cut — which points at the eye's fixed angular limits as the more durable half of the mechanism.

Where does that leave the perceptual and the conventional? On both sides, and the split is worth being precise about.

Perceptual and physical: that a fixed absolute imaging error is ruinous at small size and negligible at large is arithmetic, and that crowding and acuity are angular limits a reader carries to every size is measurement, not taste. These parts would hold for someone who had never seen a poster.

Conventional: the reading of an enlarged caption face as "clumsy" rather than "sturdy" is learned. Centuries of display cutting have trained us to expect high contrast and tight fit at large sizes, so a face lacking them registers as provincial rather than merely robust — and a design movement can invert that deliberately, since heavy, low-contrast display type has been fashionable more than once. The constraint at small sizes is close to a hard one; the disappointment at large sizes is soft, and rests on expectation.

Two boundaries against neighbours. The legibility question asks which typographic variables speed decoding at a given size, and answers with x-height, spacing and line length. The typeface-voice question asks what the shapes connote, and answers with the company a face has kept. This is neither: it asks why the optimal setting of those variables moves with size, so that no single drawing can be right at both ends — which is also why the three answers are compatible.

• How crowding specifically limits letter identification, and why it bites harder away from the point of fixation.

Every term the collection defines is gathered in the glossary.