For the determined reader
How this was made
Every image in this exhibition that is not a photograph or an old plate was computed, and every number was checked by a test written separately from the page that shows it. This page collects the methods, the data, the tests and the limits in one place.
The method
Each chapter follows the same path. It opens on a question and asks you to commit to an answer before you see one, because a guess you have made is a guess you can learn from. It then shows what happens, and spends a station on why the wrong answers were reasonable, since most wrong answers are right about something. A bench lets you change the conditions and see the rule at work. A case you have not seen tests whether the idea transfers, and is chosen so that a memorised answer gives the wrong result. The chapter ends with the history, which is usually a dispute, and with where the explanation stops. The epilogue gathers the histories into one long argument about what light is, asks you to explain it back, and ends with what nobody knows yet.
Underneath each chapter is a physics engine, a small program that computes what the page shows. The same engine runs in your browser and in the tests. Each test file reimplements the physics independently, usually in a different language, compares it with the shipped engine, and checks every number the page states. Where a measurement exists, the tests check against it. Where the writer guessed a result before computing it, the guess is recorded, including the ones that were wrong.
Chapters, engines and checks
| Chapter | What is computed | Checks | The check that matters most |
|---|---|---|---|
| 1 · The Pinhole | Fresnel diffraction through a round hole; blur against hole size | 14 | The sharpest hole, about 1.5√(λf), found by brute force and by formula |
| 2 · What the Objective Caught | Abbe imaging of a diatom with partially coherent light | 38 | The striae vanish when the objective cannot catch a first diffraction order, and appear with oblique light |
| 3 · In Step | Fringe contrast from extended sources (van Cittert–Zernike) | 21 | Strip and disc formulas against a direct sum over points of the source |
| 4 · The Black Cross | Jones calculus; interference colours through the CIE 1931 observer | 27 | Matrix products against closed-form intensities; the Michel-Lévy sequence |
| 5 · One at a Time | Single detections drawn from computed patterns | 15 | Noise growing as √N; detections needed rising as 1/V² |
| 6 · The Last Instrument | Frequency of seeing from Poisson statistics and dark events | 18 | The classic seeing curves, and the false-alarm rate when you decide to see more readily |
| 7 · The Thermometer Beyond Red | Sunlight spread by flint and crown prisms; heating per unit of table | 25 | Sellmeier coefficients reproduce Schott’s catalogue indices to five decimals; the peak stays past the red with real sunlight |
| 8 · The Window | How far light of each wavelength travels in pure water | 20 | The shipped data against the raw tables and three independent measurements |
| 9 · Windows in the Sky | The direct beam through ozone, air and aerosol; sunburn weighting | 19 | The model, given the standard’s assumptions, reproduces the ASTM G173 spectrum within a few percent |
| 10 · The Size of a Dish | Sidereal drift; diffraction-limited beams from an eye to the Event Horizon Telescope | 21 | Airy constants re-derived from the Bessel function; beams against published instrument figures |
| 11 · A New Kind of Ray | X-ray attenuation in flesh and bone; a computed radiograph | 19 | NIST’s element tables, mixed by weight, reproduce NIST’s own bone and tissue within 0.7% |
| 12 · The Hardest Light | Compton scattering, Klein–Nishina, a gamma-ray spectrum built count by count | 19 | The Compton formula against a direct solve of energy and momentum; Klein–Nishina against NIST’s measured water to 0.05% |
Measured data
- Glass. Schott optical glass catalogue, Sellmeier coefficients for F2 and N-BK7 (chapter 7).
- Sunlight. ASTM G173-03 reference spectra, generated with SMARTS 2.9.2: above the air, and the direct beam at air mass 1.5 (chapters 7 and 9).
- Ozone. A. Serdyuchenko and colleagues, Atmospheric Measurement Techniques 7 (2014), absorption cross-sections at 293 K, via the MPI-Mainz UV/VIS Spectral Atlas (chapter 9).
- Water. R. M. Pope and E. S. Fry, Applied Optics 36 (1997), 380 to 727.5 nm; D. J. Segelstein (1981) elsewhere; both as tabulated by S. Prahl (chapter 8).
- X-ray attenuation. J. H. Hubbell and S. M. Seltzer, NIST Standard Reference Database 126, with ICRU-44 tissue compositions (chapters 11 and 12).
- Colour vision. The CIE 1931 standard observer, through the analytic fit of Wyman, Sloan and Shirley (2013) (chapters 4 and 7).
- Constants. CODATA 2018 values for the electron, Planck’s constant and the speed of light; IAU values for the sidereal day.
The raw files are kept with the source, alongside the processed versions the pages load, so every number can be traced back.
What the exhibition does not claim
Each chapter ends with where its explanation stops, and each model says what it leaves out. The largest simplifications, collected:
- Light is treated as a scalar wave in the first three chapters, and polarisation enters only in chapter 4.
- The ideal Sun of chapter 7 is a 5772 K black body; the bench offers real sunlight for comparison.
- Chapter 9 models only the direct beam of sunlight; the sky’s scattered ultraviolet is described but not computed.
- X-ray tubes give a spread of energies; chapter 11 uses a single effective energy, and its forearm is a sketch.
- The full-spectrum strip in chapter 9, the Moon in chapter 10 and the detector in chapter 12 are labelled as schematics, and the physics inside each is exact.
- History is summarised from primary papers where they could be read, and from standard secondary accounts where not. The chapter notes list what still needs checking against sources.
Images
The historical plates are from the Wellcome Collection and are in the public domain, except the photograph of starch grains in chapter 4 by Lauren Holden (CC BY 4.0) and Ernst Haeckel’s diatoms in chapter 2 (1904, public domain). Each is credited in its caption. Images marked computed were drawn by the chapters’ own engines; images marked schematic illustrate an idea without claiming to be a measurement.
The working files
The notes behind each chapter, with its open questions, and the outputs of its tests:
The notes are plain text with light formatting, written for the record rather than for display.
Privacy
The exhibition sets no cookies, runs no analytics and loads nothing from any other website: its fonts, images, data and code are all served from here. Your predictions and your notebook are kept in your own browser’s storage and never leave your device. The “Clear notebook” button on the front door erases them.
Who made it
The Shape of Light was conceived and directed by Daniel Turrietta. The text, the physics engines, the tests and the pages were drafted with Claude, an AI model made by Anthropic, and revised under Daniel’s direction. The tests exist because a model, like a person, can be fluently wrong; every number on these pages has been checked by one.
Back to the exhibition