Fedderson-Daedalus: PhD mathematician review (Step 2)

Role: checks derivations, units, hidden assumptions and arithmetic. Source tags: [read] = from the document, [read: page image p.N] = read from rendered page images, [web] = URL, [memory] = recall with confidence.

Source gap (repeated from Step 1). Both PDFs are Google Books scans with no text layer. Everything below was read from rendered page images. The "Daedalus" file is the Proceedings of the American Academy of Arts and Sciences vol. 25 (1890), not Feddersen's own papers and not the journal Daedalus. Feddersen's 1857-1862 papers are not in the folder; statements about them rest on the 1890 authors' summary and on [memory]/[web]. Unlike Step 1, I read every page of Rich (pp. 140-155 of the reprint, PDF pp. 6-24) and the whole Trowbridge-Sabine paper (book pp. 109-123) plus Trowbridge pp. 192-194. Still not read: the other ~380 pages of the Proceedings. No independent replication of Rich was searched for successfully: the web search quota was exhausted, so "no replication found" below means "not checked", not "none exists".

Corrections to Step 1

1. Formula. Step 1 gives t = 2 pi sqrt(LC) as the formula tested. The Trowbridge-Sabine paper states it two ways. On p.109 it is the bare t = 2 pi sqrt(LC) (Hertz) [read: page image p.109]. On p.110, where Lorenz's comparison is described, it is t = (2 pi / v) sqrt(LC), with v = 300 x 10^6 m/s the ratio of electrostatic to electromagnetic units [read: page image p.110]. The 1/v is needed because C is in electrostatic units (cm) and L in electromagnetic units (cm). Their own numbers need it: C = 5317.9 cm and L = 39,700 cm give t = 2 pi sqrt(39700 x 5317.9) / 3.0e10 cm/s = 3.045 microseconds, matching the paper's "theoretical time .00000304 s" [read: page image p.115; my calculation]. The Step 1 shorthand is correct only in SI units.
2. Rich's polarity ambiguity is resolved, and it is worse than Step 1 thought. Step 1 said the convention was merely ambiguous. In fact: for Cu-Fe, polarity was measured with a DC make/break test and a ballistic galvanometer, and the half-cycle with iron NEGATIVE starts the spark more readily (Cu+Fe-: 6.2 or 6.8 trains, vs 3.16) [read: page images pp.150-152]. The "more readily from iron" language in the prose is tied to "iron anode" in some places (p.151 on vapour light; p.153 Fe-Zn text) and to "iron negative" in others (p.151 Fe-Bi text). The Fe-Zn and Cu-Bi tables contradict their own accompanying text (see claim 5). Step 1 wrote that Cu+Zn- has only 11 half-cycles; true, but see claim 5 for what that does to the average.
3. Arithmetic slip in Rich's Cu-Fe summary (not in Step 1). The table says 30 half-cycles, 205 trains, "average 6.2". 205/30 = 6.83. I transcribed the full "typical CuFe" record on p.147 (30 complete half-cycles with Fe negative, 32 with Fe positive): the train counts sum to exactly 205 and 101, so the totals are right and the printed 6.2 is a slip (6.2 would need 33 half-cycles). The asymmetry is therefore larger than printed (6.83 vs 3.16).
4. Trowbridge's sensitivity. Step 1 says "sensitivity of about 1/100 of a wavelength". The printed fraction is hard to read (it looks like 1/100, possibly 1/1000), and the unit is only "a wave length". Taken literally as a fraction of lambda (about 40 Angstrom at 4000 Angstrom) it would be far too coarse to see the predicted shift (about 0.5 Angstrom). The text also says the predicted broadening "might amount to a whole tenth meter"; a "tenth-metre" is the old name for the Angstrom (10^-10 m), so the prediction is about 1 Angstrom, which fits the calculation below. I cannot determine the intended sensitivity from the scan [read: page image p.193].
5. Date of Feddersen's oscillation proof. Step 1 says 1857-1862 and "confirmed Kelvin's 1853 prediction" [memory, high]. [web] https://en.wikipedia.org/wiki/Berend_Wilhelm_Feddersen says 1859 for the demonstration of oscillatory spark discharge and does not mention Kelvin. The Trowbridge-Sabine footnote cites Feddersen in Poggendorff's Annalen vols. 103 (1858), 108 (1859), 112 (1861), 113 (1861), 115 (1862), 116 (1862) [read: page image p.109]. The Kelvin 1853 link remains [memory, high], not web-checked (search quota exhausted).

Everything else in facts.json and claims.md matches what I read. The Step 1 table figures all re-add correctly (checked train/spark counts for CuFe, CuCu, FeBi, CuBi, FeZn, CuZn).

---

Claim 1. A condenser discharge through a spark gap is oscillatory with period t = 2 pi sqrt(LC), visible in a rotating mirror (Feddersen; Trowbridge-Sabine background)

a. Mathematics. Consistent. The series RLC circuit gives angular frequency w = sqrt(1/(LC) - R^2/(4L^2)), so the period is 2 pi sqrt(LC) only when R << 2 sqrt(L/C). Check with the paper's own circuit: L = 39.7 microhenry (1 cm = 1 nH), C = 5317.9 cm = 5.92 nF, so the characteristic impedance sqrt(L/C) = 82 ohm. The wire resistance of 1.54 ohm gives damping ratio R/(2 Z0) = 0.0094 and a period lengthening of only 4 x 10^-5 [my calculation]. So the formula is accurate for the wire alone; the spark's own resistance is the unknown. Capacitance formula C = l / (2 ln(b/a)) for coaxial cylinders is the correct CGS expression [read: page image p.111]. For two parallel wires, L/l = 2 ln(b^2/a^2) + 1 = 4 ln(b/a) + 1 in EMU is the Maxwell form; the "+1" is the internal-inductance term; the quoted Rayleigh high-frequency correction is the skin-effect modification [read: page images pp.113-114]. Units: the authors use mixed ESU/EMU and hide the factor v inside the formula, which is easy to misread, but the paper's numbers are consistent once v is included. b. Experiment. Consistent: damped LC oscillation is textbook and was confirmed by Hertz's experiments (1887-88) [memory, high]. The measured periods in the paper's tables are 2.9-3.4 microseconds against a computed 3.04 [read: page images pp.115-116], i.e. within about 10%. c. Tested. Yes: Feddersen (rotating mirror, Annalen 1857-1862), Lorenz (1879), Hertz, and Trowbridge-Sabine, with further oscilloscope-era confirmation [memory, high for Feddersen/Hertz; Lorenz per the paper's own citation]. Replicated many times. d. If true. Nothing changes; it is the mainstream picture. Verdict: Established.

Claim 2. Measured periods deviate from the LC formula, attributed to "dielectric viscosity" and energy transformations in the air dielectric (Trowbridge-Sabine 1890)

a. Mathematics. Weak. Three points. (i) The deviation is never expressed as a number with an error. The paper plots successive periods against the order of the oscillation (Figs. 1-6); successive values scatter between roughly 2.9 and 3.4 microseconds, with the first half-oscillation (doubled) near 3.3 [read: page images pp.115-117, 119]. (ii) The explanation is by analogy (hysteresis in magnetism; Laplace's heating-cooling correction to the speed of sound) with no equation: no relaxation time, no loss function, no prediction that could be compared to the data [read: page images pp.120, 122]. (iii) The alternative they dismiss, a varying ohmic resistance of the spark path, is rejected with the sentence "inadequate to explain the whole effect" and one test with a glass jar [read: page image p.120]. Size of that alternative: a period lengthening of 5% needs a series resistance of about 50 ohm, 10% needs about 68 ohm (from T = T0 / sqrt(1 - (R/2Z0)^2), Z0 = 82 ohm) [my calculation]. A spark channel of that resistance in the first few hundred nanoseconds of breakdown is plausible [memory, medium; not web-checked], and a resistance that falls as the channel heats would give exactly a time-varying period. The authors also reasonably argue that measurement error would give alternately long and short periods, which is not seen [read: page image p.120]; that is a fair check on random error but not on systematic physics. b. Experiment. The existence of period shifts in spark circuits is consistent with later work (nonlinear spark resistance, stray L and C) [memory, medium; not web-checked]. "Dielectric viscosity of air" is not an accepted mechanism: air at these frequencies has no measurable dielectric relaxation [memory, medium; not web-checked]. c. Tested. Only by the authors; no replication [web check not possible: quota exhausted; none known]. d. If true. Air would need a frequency-dependent permittivity at megahertz frequencies, contradicting the near-constant value (about 1.0006) used throughout; nothing else would change. Verdict (split). The observation of period scatter of order 5-10% in a spark circuit: Plausible, untested (as to a neutral replication; it is consistent with known physics). The explanation "dielectric viscosity": Historically superseded but instructive (it is a reasonable 1890 analogy with no testable form; spark resistance is the better candidate and was not excluded).

Claim 3. In an oscillatory spark the metal atoms are not carried to and fro; they are only shaken so they radiate (Trowbridge 1890)

a. Mathematics. The Doppler formula lambda1 = lambda V/(V + d) is first-order correct (V = speed of light, d = speed of approach) [read: page image p.193]. Check of the prediction: gap 6 mm, half-period 1.5 x 10^-7 s for t = 3 x 10^-7 s, so v = 0.6 cm / 1.5e-7 s = 4 x 10^4 m/s, Doppler shift v/c = 1.3 x 10^-4, about 0.5 Angstrom at 4000 Angstrom, so about 1 Angstrom of broadening, matching "a tenth meter" [my calculation]. For t = 2.4 x 10^-6 s the speed is 5 x 10^3 m/s and the shift 0.07 Angstrom. Hidden assumption: that an iron atom crosses the entire gap each half-cycle. Energy check: an iron atom (56 u) at 4 x 10^4 m/s has kinetic energy 0.5 m v^2 = 460 eV [my calculation]. Thomson's argument, quoted in the paper, is the same energy-budget argument for the speed of discharge propagation [read: page images pp.192-193]. At atmospheric pressure collisions limit drift: ion mobility in air is of order 1-2 cm^2/(V s) [memory, medium; the web check failed to find a figure], so even in 3 x 10^6 V/m the drift is about 10^2-10^3 m/s, which would be a shift of 10^-3 to 10^-2 Angstrom, below the grating's reach. So the test (null) is expected, but it tests a hypothesis that was already excluded by an energy budget, and so is weaker than it looks. b. Experiment. Consistent with all later spectroscopy: line broadening in sparks is dominated by Stark and thermal (Doppler) broadening, not coherent oscillation [memory, high]. c. Tested. One experiment, by Trowbridge, no quantitative sensitivity or line-width data given, one photograph pair; no replication on record that I could check. d. If true. Nothing; it is mainstream. Verdict (split). The statement that atoms do not oscillate across the gap in step with the field: Established (by the energy and collision-limited drift calculation above; Trowbridge's null is consistent but is a single unreplicated, unquantified photograph). The stronger statement in the paper that his experiment "strongly" shows this: not independent evidence, since the hypothesis was never realistic.

Claim 4. A spark between unlike metals is initiated more readily on one polarity, so the gap partly rectifies (Rich 1917)

a. Mathematics. The counts are arithmetically sound apart from the Cu-Fe average slip (correction 3) and the Cu+Zn- average (claim 5). I re-added the spark totals for CuFe, CuCu, FeBi, CuBi, FeZn, CuZn from the train-length histograms; all match. Statistics: for the transcribed Cu-Fe record the train counts per half-cycle are mean 6.83 (SD 1.90, n = 30) against 3.16 (SD 0.81, n = 32); the Welch t statistic is 9.8 [my calculation]. That is large, but half-cycles in one film are not independent (drift in gap and charge), and there is one film per metal pair, so there is no estimate of film-to-film or electrode-to-electrode variance. Control: the Cu-Cu run gives 2.03 vs 2.24 trains per half-cycle and 3.97 vs 3.93 sparks per train, supporting symmetry of the supply [read: page image p.149]. This control matters because Rich himself notes the leaky-transformer primary flux alone would bias one half-cycle [read: page image p.145]. However the claim "in every photograph, without a single exception" the Cu-Cu was symmetric is not backed with data beyond one film. b. Experiment. Polarity dependence of breakdown and of arcs is real physics: point/plane breakdown in air differs by polarity, and metal-oxide-metal contacts rectify (copper-oxide rectifier, Grondahl, patent 1927, in industrial use from 1924) [web] https://en.wikipedia.org/wiki/Copper(I)_oxide. Rich's electrodes were bare Cu, Fe, Zn, Bi in air, so surface oxide films, sharpness and heating asymmetry are uncontrolled confounders [reviewer judgement]. At gaps 0.1 to 4 mm the dependence on cathode material in Paschen's law enters only logarithmically through a secondary-emission coefficient [memory, high]; Rich's earlier-cited null results (Righi, Peace, Carr) were at larger gaps with DC [read: page images pp.140-141]. c. Tested. Only by Rich (University of Michigan dissertation, 1915; published 1917) under K. E. Guthe. No replication was found; I could not check citations (web search quota exhausted). The nearest independent evidence is Guthe's 1901 coherer polarity effect, which Rich cites [read: page images pp.141, 154-155]. d. If true. Nothing fundamental; it is a refinement of electrode-surface physics already in the standard picture (work function, cathode fall, surface films). Verdict: Never properly tested. A qualitative asymmetry for Cu-Fe is plausible and was measured with a sound polarity check; there was no neutral, controlled replication on record, and not dismissed without testing, simply not followed up as far as I could find.

Claim 5. The ordering Fe, Bi, Zn, Cu for easy initiation, interpreted as ease of electron emission; electrode material is therefore not negligible (Rich 1917)

a. Mathematics. This is where the paper's internal logic fails. - Tables vs text. Reading the headings: Fe-Bi+ (Fe negative) 5.21 trains vs Fe+Bi- 2.86 [p.152]; Fe-Zn+ (Fe negative) 3.1 vs Fe+Zn- 1.41 [p.153]; Cu+Bi- (Bi negative) 2.16 vs Cu-Bi+ 1.06 [p.152]; Cu-Zn+ (Zn positive) 4.77 vs Cu+Zn- 1.27 [p.154]. The text beside these tables says: Fe-Zn, "half-cycles in which iron was initially the anode produced considerably more than twice as many trains" (the heading puts Fe negative); Cu-Bi, "twice as many spark trains originated from the bismuth anode as from the copper anode" (the heading has Bi negative giving 2.16 and Bi positive 1.06, the opposite); Cu-Zn, zinc anode more productive (heading agrees) [read: page images pp.152-154]. So in two of the four tables the headings and the prose assign the opposite polarity. - Which series do the headings give? Treating the headings as correct and the lower-trains electrode as the harder emitter: Fe > Cu, Fe > Bi, Fe > Zn, Bi > Cu, but Cu > Zn (Cu negative is the better half-cycle in Cu-Zn). That contradicts the stated order Fe, Bi, Zn, Cu (which has Zn above Cu). Treating the prose as correct requires reversing the Fe-Bi and Cu-Fe statements that are anode-based in other places. No single reading makes all six pairs agree with the series. - Sign convention. Rich writes "always assuming that the discharge is electronic, so that the current flows from anode to cathode" [read: page image p.152] and calls the summary series "electrons are emitted from iron more easily" [p.155]. Electrons leave the cathode (negative electrode) and travel to the anode, opposite to conventional current. If the prose "anode" statements are meant as emitters, they have the sign backward. - Averaging. In the Cu-Zn table the 1.27 for Cu+Zn- is 14 trains over 11 half-cycles (his footnote), excluding the 21 empty half-cycles of 32 observed; the other column is 143/30 = 4.77 over all half-cycles. Averaged the same way Cu+Zn- is 14/32 = 0.44, so the asymmetry is more extreme than the printed numbers, but the two columns are not computed consistently [read: page image p.154; my calculation]. - Cu-Fe average is misprinted (correction 3). - Odd/even count. In Cu-Cu 71.5% and 86.5% of trains are even, in Cu+Fe- 92%; so even-dominance is a base-rate feature of this circuit, and the 82% odd in Cu-Fe+ is the real departure. Rich himself says the number-of-sparks measure cannot be relied on [read: page image p.154]. b. Experiment. The series does not follow the order of work functions. [web] https://en.wikipedia.org/wiki/Work_function gives Fe 4.67-4.81 eV, Cu 4.53-5.10 eV, Zn 3.63-4.9 eV, Bi 4.31 eV (ranges vary with crystal face and surface state). By work function alone Zn and Bi should be the easiest emitters and Fe and Cu similar; Rich's finding that iron is best and Zn is no better than Cu is not what electron emission predicts. In atmospheric-pressure sparks, initiation is controlled by the avalanche and streamer mechanism and by surface films and field enhancement at microscopic points, not primarily by thermionic or field emission from a clean metal [memory, high for the avalanche picture]. c. Tested. Only by Rich; Guthe's coherer series (Bi, Fe, Zn, Cu) differs by the positions of Bi and Fe; the coherer is a different (contact) regime [read: page images pp.154-155]. No replication found. d. If true. If the series were really governed by electron emission, spark thresholds in air would have to depend strongly on the cathode metal, requiring a revision of the picture that cathode material enters only through a logarithmic secondary-emission term. As stated the data are too inconsistent to force any change. Verdict (split). "Electrode material is not negligible": Plausible, untested (as in claim 4). The specific series Fe, Bi, Zn, Cu with its electron-emission interpretation: Internally inconsistent (headings contradict the prose in two of four tables; the series is not reproduced by the headings for Zn vs Cu; the anode/emitter sign is backward), and in addition not supported by work-function data.

---

Summary table

Files: review/Fedderson-Daedalus/mathematician.md