The Mechanism Seam · a reference conjured from no reference

Flat Out of Nothing

To check that a surface is flat you hold it against something already flat. So where did the first flat come from? There was no truer reference to borrow. The answer is a bootstrap hiding in Victorian shop practice: work three plates against each other in pairs, and flatness appears out of nothing.

Two plates worked face to face will mate: the high spots wear until they touch everywhere. Engineer's blue transfers across the whole face, the fit reads perfect, and yet nothing forces them to be flat. One can be a shallow dome and the other its exact bowl, sharing some radius R that no test between the two can pin down. Watch that trap first.

That is the whole trap in one line. Lapping drives the pair to the mating condition a(x) + b(x) = constant: wherever their combined height stands proud, contact wears it down, until the gap is zero. If plate A settles into a dome of curvature +1/R, plate B is forced to the exact bowl -1/R. Their sum is flat; neither plate is. Any R satisfies that single equation, so a two-body comparison has a whole one-parameter family of false passes.

The third plate kills the free parameter

Bring in plate C and test all three pairs. Each mating imposes its own equation on the shared curvature:

Two of those equations still leave a free parameter: satisfy a+b=0 and b+c=0 and you have only shown a = c = -b, any value at all. The third equation c+a=0 then demands -b + -b = 0, so b = 0, and with it a = c = 0. The determinant of that little system is 2 (not zero), so the only surface that mates all three ways at once is the one with 1/R = 0: the plane. Run the lapping and watch the three curvatures fall to it.

That convergence is the invention. Joseph Whitworth probably did not discover it alone: in Henry Maudslay's London workshop the standard planes were made three at a time, rubbed each on each and finished with scrapers (James Nasmyth, who arrived there in 1829, called the art a very old mechanical dodge; Whitworth's obituaries credit the discovery to Whitworth himself, made while he worked there). What Whitworth did was systematize it and publish it, in an 1840 paper read to the British Association at Glasgow and reprinted in his collected papers as A Paper on Plane Metallic Surfaces or True Planes. (The exact title varies by printing: the British Association's own Report of the 1840 meeting abstracts it as On producing True Planes or Surfaces on Metals; Grace's Guide reports it as Plane Metallic Surfaces, and the Proper Mode of Preparing them, and the 1841 Mechanics' Magazine printing carries that wording too.) In it he argued for finishing by hand scraping rather than grinding, so that the high spots shown by the colouring matter could be taken down one at a time. Whitworth used red ochre and oil for that transfer; Prussian-blue engineer's blue is the later equivalent. On grinding he is blunt: While grinding is universally regarded as indispensable to a finished surface, it is, in fact, positively detrimental. The three-plate method needs no prior flat reference. It is the answer to the bootstrapping question of the first flat surface.

The laser still mostly leans on the bootstrap

The tidy dismissal: charming Victorian trick; lasers solved flatness. They did not. A Fizeau interferometer does not measure a flat against the truth. It measures a flat against its own reference flat, and reads out the gap. To learn the reference you need an absolute test. One, older than the laser, compares it with a liquid surface levelled by gravity (in 2002 Germany's national primary standard for flatness was a liquid mercury mirror). But the usual route is to compare flats against each other: calibration is again three flats measured pairwise, and full-aperture three-flat tests have largely superseded the liquid mirror. Schulz and Schwider set this route out in 1967, naming the liquid surface as the alternative, and its three-body structure is Whitworth's exactly, moved from microns to nanometres.

Here is the interferometric version. Along the one line about which a flat is flipped to face its partner, the coordinate is unchanged, so the three comparisons are clean sums A+B, B+C, C+A. That 3-by-3 system solves, and it recovers each flat's absolute profile along that diameter. Reveal it:

Solving that diameter is the good news. The bad news, and the reason absolute flatness stayed hard into the laser age, is the second tab. Off the flip axis the flip mirrors the surface, x → -x, so the comparison reads A(x) + B(-x). Add the same odd bow to all three flats and every measurement is untouched: the shared odd component sits in the null space of the whole test. The classical three-comparison test recovers the absolute profile along one diameter and no more. Pinning the rest takes comparisons in further positions: Schulz and Schwider's own 1967 method used a fourth interferogram to reach any diameter you choose, Fritz turned the flats and fit Zernike polynomials to the whole surface in 1984, and the N-position and shifted-map methods since do the same. Every one of them is another mutual comparison. None escapes the bootstrap; they only close the null space mode by mode.