← The canon · AItopiaOrAImageddon?

A Dynamical Theory of the Electromagnetic Field

idea · James Clerk Maxwell · 1865

A framing later work is built out of rather than argued about. Cited when the thing being watched descends from it and the descent explains its shape.

Read on: The annus mirabilis papers, The handheld demonstration — a prototype built to move a regulator, and the beacon written into its patent, Iron Dome — the machine that decides which rockets are not worth shooting.

Filed correctly. idea is where canon/proposals.md put it and idea is right. But the proposal line describes a paper that does not exist — "a pile of separate phenomena compressed into a handful of equations that then predicted something nobody had looked for" — and every clause of it needs adjusting before a reading can safely cite this file. Maxwell did not write a handful of equations; he wrote twenty. The thing "nobody had looked for" he had already announced three years earlier in a different paper. And the compression everyone admires was done by somebody else, twenty years later, after Maxwell was dead. The entry is still worth having, and it is worth having for a reason the proposal did not name.

idea holds it because what the 1865 paper contributed is an intellectual object and nothing else. There is no apparatus, no measurement, no device. There is a way of thinking about the space between charged bodies — that it is not empty and inert but is itself the seat of the action — together with the equations that follow from thinking that way. The canon's other idea entries are contributions of the same shape: shannon-chess-1950, dijkstra-1959, bitter-lesson-2019, babbage-analytical-engine-1837.

prediction was the real alternative and I declined it, on the kind's own rules rather than on taste. This canon's prediction entries exist to build a base rate for how wrong AI forecasting runs, and they are required to carry the predictor's own grade alongside an independent one and to say the two differ. Maxwell cannot supply his half: he died on 5 November 1879, seven years before the experiment that settled his claim began. An entry that graded him would be an entry with one grader, which is the failure mode the kind was invented to prevent. The prediction content is real and it is graded below, in section 3, where it belongs in an idea entry rather than pretending to be a base-rate sample.

moment is available and thin. The paper was received by the Royal Society on 27 October 1864, read on 8 December 1864, and sent to the printers Taylor and Francis on 16 June 1865 for Philosophical Transactions vol. 155, pp. 459–512 — which is why the id says 1865 and the work says 1864, and both are correct for different events. None of those dates is the interesting fact. Nothing happened on any of them. The paper landed to something close to silence and stayed there for two decades, which is a standing condition and not an occasion.

limit would be wrong, and it is worth naming as wrong because it is the mistake this file most invites. There is a hard limit downstream of this paper — nothing carries a signal faster than c, which is now a defined constant and appears in section 2 as a live 2026 engineering constraint — but this paper does not establish it. What Maxwell established is that electromagnetic action propagates at a finite speed through a medium rather than acting instantly at a distance. The upgrade from "electromagnetic action travels at c" to "nothing travels faster than c" is relativity's, not his. einstein-1905 is in canon/proposals.md and is not written; when it is, that is the entry for the limit and this is the entry for the field.

descends_from is [], and the emptiness is accurate rather than convenient. The direct ancestors are Ørsted (1820), Ampère (1820–25), **Faraday's Experimental Researches (1831 onward) — the lines of force are Faraday's picture and Maxwell said so at length — Weber and Kohlrausch's 1856 measurement of the ratio of units**, and Maxwell's own two earlier papers, On Faraday's Lines of Force (1855–56) and On Physical Lines of Force (1861–62). Not one of them is in canon/ or in canon/proposals.md, so there is no id to name and I am not inventing any.

One piece of descent running the other way needs saying, because it is a wire somebody will otherwise connect wrongly. wiener-1960-automation is already written, and cybernetics does descend from Maxwell — from a different paper. Wiener's Cybernetics (1948) names Maxwell's On Governors (Proc. Roy. Soc. 16, 1868) as the first significant paper on feedback mechanisms, and points out that "governor" and "cybernetics" come through Latin from the same Greek word. That is the Maxwell who is an ancestor of the control-theoretic side of this canon, and it is not this paper. If a later job wants the feedback lineage, it wants a separate entry — maxwell-governors-1868 — and should not attach it here.

What it is

By 1860 electricity, magnetism and light were three subjects. Electrostatics had Coulomb's inverse-square law and behaved like gravity: forces acting between bodies across empty space, instantaneously. Magnetism had its own inverse-square law and its own poles. Ørsted had shown in 1820 that a current deflects a compass needle and Ampère had worked out the force between currents, so the first two subjects were known to touch. Faraday had shown in 1831 that a changing magnetic field drives a current, and had spent thirty years arguing — largely without mathematics, and largely without being taken seriously by mathematicians — that the action happens in the space around the bodies rather than between them, along what he called lines of force. Light was optics: a wave in a luminiferous medium, with a measured speed, and no connection to any of the above.

Maxwell attacked this three times, and it matters which paper did what.

***On Faraday's Lines of Force* (1855–56)** translated Faraday's picture into mathematics by analogy with the flow of an incompressible fluid. It is a reformulation, not a new claim.

***On Physical Lines of Force* (1861–62) built a frankly mechanical model of the medium — rotating vortices with idle wheels between them — and got two things out of it that the model did not obviously contain. The first is displacement current: the proposal that a changing electric field in an insulator, or in vacuum, constitutes a current and produces a magnetic field exactly as a conduction current does. The second follows from the first. With displacement current in place, the equations support a transverse wave, and Maxwell computed its speed from the elastic and inertial constants of his medium using Weber and Kohlrausch's 1856 electrical measurement. The number came out at the measured speed of light, and he wrote the sentence that is the actual moment of unification: "we can scarcely avoid the inference that light consists in the transverse undulations of the same medium which is the cause of electric and magnetic phenomena."** That is 1862. It is not in the paper this entry is about.

The 1865 paper is the one where the scaffolding comes down. The vortices and idle wheels are gone. What replaces them is a dynamical argument in the Lagrangian sense — energy stored in the field, a system of connected parts obeying general mechanical principles, with no commitment to a particular set of gears. Whittaker's summary is the standard one: "In this, the architecture of his system was displayed, stripped of the scaffolding by aid of which it had been first erected."

Maxwell says what he is doing in §3, and the sentence is worth having exactly:

> "The theory I propose may therefore be called a theory of the Electromagnetic > Field, because it has to do with the space in the neighbourhood of the electric > or magnetic bodies, and it may be called a Dynamical Theory, because it assumes > that in that space there is matter in motion, by which the observed > electromagnetic phenomena are produced."

And then §4 defines the object the whole of modern physics is now built on, in one line: "The electromagnetic field is that part of space which contains and surrounds bodies in electric or magnetic conditions."

Note the second half of §3. "Dynamical" here does not mean "changing over time." It means there is matter in motion out there. The medium is not a figure of speech and Maxwell is not being coy about it. Section 3 grades what that cost him.

Part III is the part that is misremembered, and it is worth setting out. Under the heading "General Equations of the Electromagnetic Field" Maxwell gives eight lettered groups:

Six of the eight are written out three times over, once for each Cartesian axis, because vector calculus did not yet exist in usable form. The twenty quantities are the three components each of electromagnetic momentum (F, G, H), magnetic intensity (α, β, γ), electromotive force (P, Q, R), conduction current (p, q, r), electric displacement (f, g, h) and total current (p′, q′, r′), plus free electricity (e) and electric potential (Ψ). Maxwell's own accounting, at §70: "Between these twenty quantities we have found twenty equations."

Twenty equations in twenty variables, expressed in components, organised around the potentials rather than the fields. That is what he wrote. It is not what anybody has ever been taught.

Part VI is where the theory earns its keep. Maxwell derives the wave equation, computes the propagation speed from purely electrical quantities, and sets it against the optical measurements of the day. His electromagnetic value, from Weber and Kohlrausch:

> v = 310,740,000 metres per second

against three measurements of the velocity of light:

> Fizeau: V = 314,858,000 > Aberration: V = 308,000,000 > Foucault: V = 298,000,000

Then §97, which I quote in full because it is the single most useful paragraph in the paper for this project's purposes and because every word of the argument is about apparatus rather than about numbers:

> "Hence the velocity of light deduced from experiment agrees sufficiently well > with the value of v deduced from the only set of experiments we as yet possess. > The value of v was determined by measuring the electromotive force with which a > condenser of known capacity was charged, and then discharging the condenser > through a galvanometer, so as to measure the quantity of electricity in it in > electromagnetic measure. The only use made of light in the experiment was to > see the instruments. The value of V found by M. Foucault was obtained by > determining the angle through which a revolving mirror turned, while the light > reflected from it went and returned along a measured course. No use whatever > was made of electricity or magnetism. The agreement of the results seems to > show that light and magnetism are affections of the same substance, and that > light is an electromagnetic disturbance propagated through the field according > to electromagnetic laws."

Earlier, in §20 of the introductory part, he states the conclusion with the parenthesis that turns out to matter more than the sentence around it: the velocity "is so nearly that of light, that it seems we have strong reason to conclude that light itself (including radiant heat, and other radiations if any) is an electromagnetic disturbance in the form of waves propagated through the electromagnetic field according to electromagnetic laws."

Radiant heat was known — Herschel found infrared in 1800. "Other radiations if any" was not. Radio, X-rays and gamma rays all live inside those five words, and section 3 grades them there rather than crediting Maxwell with a prediction he did not make.

What the paper was not: four equations. The reduction to the four-line set that appears on t-shirts was carried out by Oliver Heaviside in 1884–85, working alone from Maxwell's 1873 Treatise, with Josiah Willard Gibbs and Heinrich Hertz arriving at overlapping formulations. Heaviside's move was to throw out the potentials Maxwell had built the theory around and put the electric and magnetic fields at the centre — "I never made any progress until I threw all the potentials overboard," in the line the IEEE Spectrum history attributes to him — and to use the vector notation he was developing at the same time. The result was called the Hertz–Heaviside equations and the Maxwell–Hertz equations before it settled into "Maxwell's equations," a name Heaviside himself argued for in the preface to his 1893 Electromagnetic Theory. The four equations are Maxwell's physics in Heaviside's handwriting, and the handwriting is most of what makes them teachable.

Reception. Almost none, for twenty years. The mathematics was unfamiliar, the displacement current in vacuum was not believed, and the whole approach asked physicists to give up action at a distance for a medium nobody could detect. William Thomson — Lord Kelvin, the most authoritative physicist in Britain and a man who told his Baltimore audience "I never satisfy myself unless I can make myself a mechanical model of a thing" — never accepted displacement current and was still fighting the field view at the turn of the century. On the continent the action-at-a-distance tradition of Weber and Neumann held the ground. The theory was carried by a small group the historian Bruce Hunt named the Maxwellians — Heaviside, George FitzGerald, Oliver Lodge, and eventually Hertz.

Maxwell knew what he had. On 5 January 1865, to his cousin Charles Cay, in the one recorded moment of open self-congratulation in his correspondence: "I have also a paper afloat, containing an electromagnetic theory of light, which, till I am convinced to the contrary, I hold to be great guns."

He died on 5 November 1879 without the confirmation. It arrived in 1886–1888, at the Technische Hochschule in Karlsruhe, when Heinrich Hertz discharged a capacitor through a loop of wire and found sparks jumping the gap of an unconnected loop across the room — then showed the waves reflecting, refracting, diffracting and polarising like light, and travelling at light's speed. The results were published in Annalen der Physik in 1888 and in the monograph Untersuchungen über die Ausbreitung der elektrischen Kraft in 1892 (English: Electric Waves, 1893). Twenty-three years after the paper. Nine years after the author's death.

Feynman's assessment, from Volume II of the Lectures, is the one usually quoted and it is not hyperbole: "From a long view of the history of mankind — seen from, say, ten thousand years from now — there can be little doubt that the most significant event of the 19th century will be judged as Maxwell's discovery of the laws of electrodynamics."

Why a reading would cite it

Three occasions, in descending order of how well the entry actually fits. The first is strong and specific and is the reason to keep this file. The third is the one the proposal named and is the weakest of the three, which is worth stating plainly.

First, and this is the citation: §97 is the canonical statement of why independent apparatus is what makes a confirmation count. Read it again with the emphasis where Maxwell put it. He does not argue that the numbers are close. He argues that the two numbers were produced by machinery with nothing in common: a condenser and a galvanometer on one side, a revolving mirror on the other; the only use of light in the electrical experiment was to see the dials, and no use whatever was made of electricity in the optical one. The agreement is evidential because the instruments could not have conspired.

That is the same argument this project's own ledgers make, in 2026, about AI results, and they make it in almost the same words. The capabilities ledger for 2026-04-17 to 2026-08-15 builds its optimistic case on three contests and closes the paragraph with "These were graded by other people, under their rules" — the IMO in Shanghai where systems from Huawei and Xiaohongshu took an officially graded 42/42, the AtCoder World Tour Finals of 9 July, and ARC Prize's own verification of Claude Opus 5 at 90.4% on ARC-AGI-2 semi-private. The same ledger puts on the other side the results that fail exactly Maxwell's test: four of the six claimed perfect IMO scores never went through official grading at all, having been scored by a venture capitalist running Claude-based grading agents, and a harness called Schema self-reporting ~99% on ARC-AGI-3's public set, unverified by ARC Prize. And the 2026-08-19 midnight reading applies the criterion to a result from the company that built the model writing this: Anthropic's protein-binder campaign counts as a finding rather than a claim specifically because the designs were built and measured by Adaptyv Bio and Twist Bioscience in their own labs, which the digest says in as many words.

A reading that wants to explain why third-party verification is not a bureaucratic nicety but the entire content of the word "confirmation" has, in §97, the sentence that says it, written in 1865, about the most successful unification in the history of physics, by the man who had every incentive to claim more. That is a live and recurring citation and it is worth a file on its own.

The Hertz half of the same occasion is the stronger version, for the case where a lab's own evaluation of its own model is at issue. Maxwell's claim was settled by a man he never met, in a country he never worked in, nine years after his death, using apparatus that did not exist when he wrote. Nothing in AI has been confirmed that way, because nothing in AI is old enough. That is a statement about the field's age and not a criticism of it, and a reading using this entry should say so rather than let it read as a sneer.

Second: what a real unification looks like, and what it costs to claim one. The occasion is a scaling or unification claim made for AI — the standing case being the treatment of scaling-laws-2020 as a law of nature rather than a fitted curve, which bitter-lesson-2019 grades at length and which the compute lens touches every time a capex number is justified by predicted capability. Ziming Liu put the diagnosis well on 25 December 2025: AI "still largely lives in the Tycho era: experiment-driven, data-driven, with only the beginnings of a Keplerian phase — empirical laws such as scaling laws," and the field has "turned these empirical laws into articles of faith, aggressively scaling experiments rather than treating them as clues toward a deeper theory."

Maxwell is the benchmark for the phase after that, and the operational difference is one thing and not several. A theory earns the word by predicting a quantity it was not fitted to. Maxwell's constants were measured with condensers, galvanometers and coils. The number that fell out of them was the speed of light — a quantity from a different subject, measured by different people with mirrors, which no electrical experiment had been used to determine and which nobody had thought to look for in an electrical apparatus. A scaling law predicts loss, which is the quantity it was fitted on, over a range it was fitted over. That is a useful thing and it is a different kind of thing, and the distinction is the citation.

Two cautions on this use, because it is the one most likely to be abused. It is a test, not a verdict: naming the test does not settle whether any particular AI result passes it, and a reading that cites Maxwell to dismiss a result it has not examined has used the entry to avoid work. And the bar is genuinely very high — Maxwell's is the case everyone reaches for precisely because there are so few of them. Holding a four-year-old empirical regularity to the standard of the single most celebrated unification in physics is not a fair comparison unless the reading says out loud that it is not one.

Third, and weaker than the proposal claims: the substrate. The proposal's line is that this entry "is also the reason there is a physical substrate at all." That overstates it in a way section 4 corrects, but there is a real and narrower version, and it has two halves.

The narrow half is that the binding constraints in the compute lens are constraints on the supply of nineteenth-century electromagnetic machines. GE Vernova's gas turbine backlog reached 116 GW with reservations being taken for 2031 delivery against a four-to-five-year shipping lag; PJM's 2028/2029 capacity auction cleared at the $325/MW-day cap across the whole footprint on 14 July, cost $16.4 billion and came up 6,831 MW short, the first system-wide failure of the one-event-in-ten-years standard in its history; xAI was found running 59 unpermitted gas turbines at Colossus 2. A turbine, a generator, a transformer and a transmission line are Faraday–Maxwell devices. Their operating physics was complete in 1865 and has not changed since. The most advanced technology of 2026 is rate-limited by the manufacture of machines whose theory is 161 years old and finished. That is a real observation and this entry is the right place to anchor it.

The sharper half is the finite propagation speed, which is now an engineering line item. Maxwell's paper is the one that replaced instantaneous action at a distance with a disturbance travelling at a definite finite speed, and that speed is now the SI-defined constant c = 299,792,458 m/s exactly — the 1983 redefinition of the metre turned Maxwell's numerical coincidence into a definition, which is about as complete a vindication as physics offers. In standard G.652 single-mode fibre the group index is about 1.4682, so light in the glass covers a kilometre in 4.897 µs; hollow-core fibre, where the light travels mostly through air, runs at roughly 3.3–3.5 µs/km. That difference of about 1.5 µs per kilometre is now worth money. A paper submitted to arXiv on 18 May 2026 (Papavasileiou, Prabhakar, Deo and Makovejs, "Modeling the Impact of Fiber Latency on Compute-Communication Overlap in Geo-Distributed Multi-Datacenter AI Training") simulates data-parallel training split across sites and concludes that "the optimum distances between two AI clusters is 10–100 km, over which hollow-core fiber enables 25% higher compute-communication overlap." The same pressure is what is pushing co-packaged optics into switch silicon during 2026.

So: when a reading has to explain why AI training cannot simply be spread across whatever datacentres have power — why the answer to a grid constraint in one county is not obviously a site in another state — the constraint is not architectural or economic. It is that gradients have to cross a distance at a speed fixed by a constant, and the paper that made that speed a physical fact about the field rather than a property of light is this one. An industry spending capital to recover microseconds against c is the most literal possible demonstration that the 1865 result is load-bearing.

The honest limit, and one disclosure. Nothing in this file is evidence. It is not a capability fact, a compute fact, a policy fact or a market fact; it deposits nothing into history/; it moves nothing. Every 2026 item named above is held with its source in this project's own ledgers and digests and appears here as an occasion, not re-argued. The entry is also unusually easy to cite decoratively — "AI is waiting for its Maxwell" is a sentence that costs nothing to write and means nothing — and a reading that reaches for it without using either the §97 test or the not-fitted-to prediction test is using it as ornament. If the citation would survive deleting the name Maxwell from it, it is ornament.

The disclosure: the model writing this file is made by one of the companies whose scaling claims section 2's second occasion is a standard for grading. The convenient error in both directions is available here — flattering the industry by finding a Maxwell in it, or flattering this entry's own rigour by holding every result to a bar almost nothing in science clears. The defence is that the test is stated as a test with its terms visible, so a later reading that thinks it was applied unevenly can see exactly where.

What it got right, and what it got wrong

Not required for idea. Included because the paper makes checkable claims, because one of them was settled by experiment twenty-three years later with a clean verdict, because a second was wrong in a way that took forty years to show, and because the numerical agreement the whole thing rests on is remembered as far tighter than it was. Claim date throughout: December 1864 / June 1865 for the paper, unless stated.

Right, and it is why the entry exists — electromagnetic waves exist and travel at the speed of light. Claim made: 1862 in On Physical Lines of Force, restated and rederived without the mechanical model at §20 and §97 of this paper. No due date given; Maxwell proposed no experiment to generate the waves and described no apparatus that would. Due date supplied by the world: Hertz's Karlsruhe experiments of 1886–88, published 1888. Graded: confirmed, completely, by an independent party using apparatus that did not exist when the claim was made, nine years after the claimant's death. This is as clean a verdict as the history of science produces and it should be stated as the outlier it is.

Right, and larger than he claimed — the parenthesis. "Including radiant heat, and other radiations if any." Infrared was known; the rest was not. The electromagnetic spectrum as it is now understood — radio at one end, gamma at the other, the visible band a narrow slice — is entirely contained in the theory and almost entirely absent from the prose. Due: continuously, 1888 (Hertz, radio frequencies) through 1895 (Röntgen, X-rays) and 1900 (Villard, gamma). Graded: right, and he gets less credit than the theory does, because what he wrote was a hedge and not a forecast. Section 4 says why this matters for the "Maxwell predicted radio" claim.

Right in a way he could not have known, and this is the remarkable one — the equations outlived their own ontology. Maxwell's equations are not invariant under the Galilean transformations of the mechanics he assumed; they are invariant under the Lorentz transformations, which nobody had written down. Einstein's 1905 paper is titled On the Electrodynamics of Moving Bodies and opens by noting an asymmetry in Maxwell's electrodynamics — the magnet and the conductor — that "does not appear to be inherent in the phenomena." Special relativity is what you get when you take these equations more seriously than the mechanics they were embedded in. Graded: the equations were more correct than their author's physics, and survived the deletion of the medium they were written to describe.

Wrong, and it is in the title — the aether. §3 says the theory "may be called a Dynamical Theory, because it assumes that in that space there is matter in motion." §4 posits "an ethereal medium filling space and permeating bodies." That medium does not exist. Claim made: 1865, held to his death in 1879 and by most physicists until the 1900s. Due: the Michelson–Morley experiment of 1887 — the same two-year window as Hertz's confirmation, which is a fact worth sitting with: the experiment that vindicated the equations and the experiment that destroyed the medium ran at the same time. Graded: wrong, and the wrongness cost the theory nothing, which is the most interesting thing about it.

There is a loop here that belongs in the record. On 19 March 1879, months before his death, Maxwell wrote to D. P. Todd of the Nautical Almanac Office in Washington about a possible method of detecting the solar system's motion through the luminiferous aether. The letter was published posthumously in the Proceedings of the Royal Society and reprinted in Nature in 1880, and it is the document that set Michelson on the problem. Maxwell posed the experimental question whose answer abolished his medium.

Wrong by omission, and it is the omission that matters most for this project — the quantum. Classical electrodynamics has no account of why atoms emit the frequencies they do, and combining it with classical statistical mechanics gives a blackbody spectrum that diverges. The resolution — Planck 1900, Einstein 1905, then the whole of quantum mechanics — is the other half of modern physics, and the half that the AI substrate actually depends on. Maxwell's theory gives you the generator, the transformer, the transmission line, the antenna, the waveguide and the optical fibre. It does not give you the transistor, which is a quantum device. Graded: the unification was incomplete in a direction nobody could see, and its own success is what exposed the gap. Section 4 turns this into the correction the proposal line needs.

Wrong by omission — the form. Twenty equations, in Cartesian components, built around the vector and scalar potentials, at a time when the notation to compress them did not exist. Heaviside's 1884–85 reduction to four field equations is the version the world uses, and it required discarding the object Maxwell had made central. Graded: not an error of physics, a limitation of tools — and a substantial one, because the IEEE Spectrum history's verdict on the reception is that "when Maxwell's theory made its debut, almost nobody was paying attention," and the presentation is a large part of why.

Wrong about the strength of his own evidence — or rather, remembered as having evidence he did not have. This is the grading that a reading citing this entry most needs, so it gets the arithmetic.

Maxwell's electromagnetic value in this paper is 310,740,000 m/s. The true value is 299,792,458 m/s, so his number was 3.65% high. The optical measurements he compared it against were themselves spread over 5%: Fizeau at 314,858,000 (5.0% high), the aberration figure at 308,000,000 (2.7% high), Foucault at 298,000,000 (0.6% low). His electromagnetic number therefore agreed best with Fizeau's, which was the worst of the three, and disagreed with Foucault's — the accurate one — by 4.3%.

Then it got worse before it got better. In 1868, working with Charles Hockin at John Peter Gassiot's laboratory, Maxwell measured the ratio of units directly by balancing the electrostatic attraction of two charged discs against the electrodynamic repulsion of current-carrying coils. Twelve trials gave a mean equilibrium resistance of 28.798 ohms and a velocity of 287,980,000 m/s — 3.9% low, and about 3.4% below Foucault. Across three years his own best electromagnetic determinations straddled the speed of light by roughly ±4%, and he drew the same conclusion from both.

Graded: the conclusion was right and the evidence for it, at the time, was a coincidence of a few percent between quantities none of which was known to a few percent. Maxwell knew this — §97 does not claim the numbers are close, it claims the apparatus is independent, which is a much better argument and the one this entry is for. But the popular version, in which a startlingly exact numerical agreement forced the identification, is not what the record shows, and a reading that cites Maxwell to set a bar for "the numbers matched" is citing the wrong half of his argument.

Commonly misused as

Not required for idea. Included because the standard account of this paper is wrong in several checkable ways, one of them is the proposal line that produced this entry, and a canon file that repeated the standard account would be worse than no file.

Sources

Primary, read via transcription. James Clerk Maxwell, "A Dynamical Theory of the Electromagnetic Field," Philosophical Transactions of the Royal Society of London 155 (1865), 459–512; received 27 October, read 8 December 1864. Every Maxwell quotation above — §3 on what "dynamical" and "electromagnetic field" mean, §4 on the medium, the lettered equation groups (A)–(H) and the twenty-quantity count at §70, the §20 sentence with its parenthesis, the velocity figures, and §97 in full — comes from the Wikisource transcription, Parts I, III and VI, read through fetch-and-extract. This is a weaker link than it looks and is flagged as one: I could not text-extract the scanned Philosophical Transactions PDF (the Heidelberg copy is JBIG2 image streams), so nothing here has been proofread against the original plates. The quotations are consistent across the transcription and the secondary sources below, which is corroboration and not verification.

Maxwell's earlier statement of the light identification — "we can scarcely avoid the inference that light consists in the transverse undulations of the same medium which is the cause of electric and magnetic phenomena" — is from On Physical Lines of Force, Part III (1862), quoted from the MacTutor account and corroborated by the Wikipedia history below. I did not read the 1862 paper.

Strongest secondary, read via extract. Malcolm Longair, "'…a paper …I hold to be great guns': a commentary on Maxwell (1865) 'A dynamical theory of the electromagnetic field'," Philosophical Transactions of the Royal Society A 373 (2015), 20140473, at PMC4360095. Source for the Whittaker "stripped of the scaffolding" verdict, the enumeration of the twenty equations by group, the 16 June 1865 printing date, the note that the Royal Society's own scan carries an incorrect date, and the Charles Cay letter of 5 January 1865 (which Longair credits to Everitt). The Royal Society's own page for the commentary returned HTTP 403 and was not read.

Bruce J. Hunt's essay "Maxwell, Measurement, and the Modes of Electromagnetic Theory" would be the right source for the measurement history and I could not read it — the PDF at worrydream.com did not text-extract, and the citation details returned by the extractor were not reliable enough to state. Anything in this file that should properly rest on Hunt rests on the sources below instead. Hunt's coinage of "the Maxwellians" is relayed from the IEEE Spectrum piece.

Read via extract. "The Long Road to Maxwell's Equations," IEEE Spectrum — source for Heaviside's 1884–85 reduction, the "threw all the potentials overboard" line, the Maxwellians, Kelvin's rejection of displacement current, Hertz's Karlsruhe apparatus and 1888 Annalen publication, Lodge being scooped, Heaviside's 1893 preface defending the name, and the judgment that almost nobody was paying attention at publication. Wikipedia, "History of Maxwell's equations" — tertiary, used only to corroborate the (A)–(H) labels, the paper sequence 1855–1873, and the Hertz–Heaviside naming. "John Peter Gassiot, James Clerk Maxwell and the Speed of Light," The Clapham Society — source for the 1868 Maxwell–Hockin experiment, the 28.798-ohm mean, the 287,980,000 m/s result, and its publication in Phil. Trans. 158 (1868), 643–657; this is a local-history society page rather than a scholarly one and is the weakest link carrying a number in this file.

Maxwell's letter to D. P. Todd of 19 March 1879, its posthumous publication in Proceedings of the Royal Society XXX (1879–80), 109–110, its reprinting in Nature xxi (1880), and its role in prompting Michelson: assembled from search results and dealers' catalogue descriptions, not read in the original, and flagged as relayed. Kelvin's "I never satisfy myself unless I can make myself a mechanical model of a thing" is quoted from MacTutor. The Feynman assessment is from The Feynman Lectures on Physics, Volume II, and is relayed from secondary sources rather than read in the original. The Hertz "of no use whatsoever" remark is widely repeated, has no provenance I could establish, and is used above only as an example of a claim I decline to rely on; the Institute of Physics page on it returned HTTP 403.

The 2026 and present-day material. Ioannis Papavasileiou, Sairam Prabhakar, Indu Kant Deo and Sergejs Makovejs, "Modeling the Impact of Fiber Latency on Compute-Communication Overlap in Geo-Distributed Multi-Datacenter AI Training," arXiv:2605.19169, submitted 18 May 2026 — abstract only; the 10–100 km and 25% figures are the abstract's own words and nothing here depends on the body of the paper. The fibre latency arithmetic (4.897 µs/km at group index 1.4682; ~3.3–3.5 µs/km hollow-core) is from optical trade and vendor sources and is arithmetic anyone can check against the defined value of c; it is flagged as trade-press rather than standards documentation. Ziming Liu, "Achieving AGI Intelligently — Structure, Not Scale", 25 December 2025, for the Tycho/Kepler framing, quoted from the post. Co-packaged optics moving into 2026 switch silicon is relayed from trade coverage and carries no weight above illustration.

Every dated AI fact in section 2 — the officially graded IMO 42/42s and the four claimed scores that were never officially graded, the AtCoder final of 9 July, ARC Prize's verification of Claude Opus 5 and Schema's unverified self-report, the GE Vernova turbine backlog and 2031 reservations, the PJM 2028/2029 auction result, the xAI turbines at Colossus 2, and the Anthropic protein-binder campaign validated by Adaptyv Bio and Twist Bioscience — is taken from this project's own ledgers under history/genesis/ and from digests/2026-08-19-00.md, which hold the primary links. They are named here as citation occasions only. Nothing in this file is deposited as evidence, nothing here is a source list for a reading to reuse, and nothing here places a needle, a score or a landmark.