Compton: review by a PhD mathematician

Reviewer role: checks derivations, notation, units, hidden assumptions. Project: Compton (single document, Compton_1923.pdf, A. H. Compton, "A Quantum Theory of the Scattering of X-Rays by Light Elements", Phys. Rev. 21(5), 483-502, May 1923).

Source gap (repeated from Step 1, now partly closed). The PDF has no text layer. In this step I rendered all 20 pages and read pp. 1, 3, 4, 5, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20 myself [read: page image p.N]; pp. 2 and 6 were not re-read (Step 1 read them). The derivation of Eqs. 1-30 and all tables and figures were therefore checked against the page images. Claims about later tests (Compton-Simon, Bothe-Geiger, Klein-Nishina) rest on web sources or [memory] as tagged. Not obtained: Compton's follow-up papers and the original Bothe-Geiger / Compton-Simon papers; I rely on secondary summaries for those.

My own checks were numerical (numpy, scratch files in /tmp/compton_m, tagged [calc]). Constants: h/(mc) = 0.024263 A, so 2h/(mc) = 0.048526 A [calc].

Corrections to Step 1

1. Energy-removed ratio is wrong in facts.json. facts.json says "total energy removed is less than Thomson's by the ratio 1/(1+2 alpha)^2". The paper (abstract p.1; Eq. 28, p.493) says the energy removed is sigma = sigma_0/(1+2 alpha). The factor (1+alpha)/(1+2 alpha)^2 is the true scattering coefficient sigma_s (Eq. 29), and alpha/(1+2 alpha)^2 is the absorption part (Eq. 30). Check: sigma_s + sigma_a = sigma_0 (1+2alpha)/(1+2alpha)^2 = sigma_0/(1+2alpha) ✓ [read: page image p.1, 11, 12].
2. claims.md claim 4: "total scattering is too high by (1+2alpha)^2/(1+alpha)" is correct if "total scattering" means the true scattering coefficient sigma_s (Eq. 29), not the total energy removed. Wording should say so.
3. claims.md claim 3 omits the observed-vs-calculated spread for the gamma rays. Table I (p.496): scattered wavelength observed / calculated: 45 deg 0.030 / 0.029; 90 deg 0.043 / 0.047; 135 deg 0.068 / 0.063 A. The 135 deg point is 8% above theory, the 90 deg point 9% below. "0.022 to 0.068" is an observation, the theory gives 0.063 [read: page image p.14]. I confirm 0.022 + 0.0485 sin^2(theta/2) gives 0.029, 0.046, 0.063 [calc].
4. Gamma-ray wavelengths were not measured by a spectrometer. They are inferred from absorption coefficients in lead (Hull-Rice value at 0.122 A, tau/rho proportional to lambda^3, minus an estimated scattering part), and the primary 0.022 A is itself an output of that method, then reused as input in Eq. 31 for the "calc" column (p.496). Step 1 says "wavelength measured via absorption in lead" but not that the calibration is partly circular [read: page image p.14].
5. The 1.10 vs 1.04 forward/backward claim does not reproduce. Eq. 27 at 0.7 A (alpha = 0.0347) gives I(40 deg)/I(140 deg) = 1.136 [calc], not "about 1.10" as printed (p.499). The Klein-Nishina energy-weighted value is 1.164 [calc, formula from memory: high]. So the prediction is farther from Barkla-Ayers' 1.04 than Compton states. Compton's ionization argument (p.499-500) is offered as the escape.
6. Two printing slips found in the paper that Step 1 did not note (see Claim 1 a and Claim 5 a): the sign in Eq. 1, and R vs R^2 in Eq. 23.
7. Fig. 7 / p.501 "beautiful agreement" is overstated and Step 1 did not examine it. The text itself says the 90 deg intensity is 0.037 I_0; Eq. 27 gives 0.070 I_0 at alpha = 1.09-1.10 [calc]. See Claim 5.
8. Abstract p.1 (not read in Step 1 in this detail) says the angular law is found "by an indirect and not quite rigid method". Step 1 captured this via p.491; fine.
9. Step 1 "pages not viewed" list is now obsolete; Step 1's transcription of Eqs. 3, 5, 31, Table-I-related figures, 0.730/0.708 and Ishino numbers was accurate.

Claim 1. A scattered quantum gives all its energy and momentum (p = h nu/c) to one electron, which recoils.

a. Mathematics. Internally consistent, with one printed sign slip. Eq. 1 (p.486) is printed as (m beta c/sqrt(1-beta^2))^2 = (h nu_0/c)^2 + (h nu_theta/c)^2 + 2 (h nu_0/c)(h nu_theta/c) cos theta [read: page image p.4, zoomed]. The recoil momentum is the vector difference p_0 - p_1, so the law of cosines for angle theta between the two rays gives minus 2 cos theta. With the minus sign, energy (Eq. 2) plus momentum give exactly Eq. 3, nu_theta = nu_0/(1 + 2 alpha sin^2(theta/2)): numerical residual 0.0 at alpha = 1.09, theta = 2.0 rad; with the printed plus sign the residual is 0.78 (not zero) [calc]. So Eq. 3 is right and Eq. 1 has a typo (or the angle in Fig. 1B is measured differently from theta). Eq. 6 for beta checked: 0.79859 from conservation laws equals 0.79859 from the printed Eq. 6 [calc]. Eqs. 7-8 (Doppler form equals Eq. 3 if beta' = alpha/(1+alpha)) reproduce identically [calc: 0.39314 both ways]. Hidden assumptions: free electron initially at rest; one-photon-one-electron process; plane wave treated as a photon with sharp momentum. Worked number: alpha = 1.09 (RaC gamma ray), theta = 180 deg gives beta_max = 2 alpha(1+alpha)/(1+2alpha+2alpha^2) = 0.82 [calc], matching p.497 ("0.82") and an effective beta = alpha/(1+alpha) = 0.52 ✓.

b. Experiment. Consistent with: the 0.730 vs 0.708 A shift (see Claim 3); Ellis beta-ray lines and the gamma-ray beta spectra (cited in the paper, I did not read them); later direct tests below. One known caveat: the scattered beam from graphite contains both a shifted and an unshifted line (bound electrons, treated by Compton only in the closing remarks, p.502). The abstract and p.485 stress "little or none" at the primary wavelength; modern data show the unshifted peak is real [memory: high; not web-verified here] and arises from the electron being bound to the atom, which the paper itself identifies as outside its scope.

c. Tested? Yes, and replicated. Compton and Simon (cloud chamber, 1925) and Bothe and Geiger (coincidence counters, 1925) showed the recoil electron and scattered photon appear simultaneously, supporting the individual-event picture and counting against the Bohr-Kramers-Slater statistical-energy proposal [web: https://en.wikipedia.org/wiki/Bothe%E2%80%93Geiger_coincidence_experiment ; https://www.nobelprize.org/prizes/physics/1954/bothe/lecture/ ; search summary confirmed both experiments independently in 1925]. Since then the Compton effect is used routinely in gamma spectroscopy and the shift formula is tested to high precision [memory: high].

d. What would have to change. In 1922: classical wave theory of X-ray scattering (Thomson scattering) had to be supplemented by photon momentum; this did happen. Today, nothing: it is textbook physics.

Verdict: Established. (The "all quanta by one electron" statement holds for free or loosely bound electrons; for tightly bound electrons coherent scattering also occurs, see Claim 6.)

Claim 2. Shift = (2h/mc) sin^2(theta/2) = 0.0484 sin^2(theta/2) A, independent of lambda_0.

a. Mathematics. Derived exactly from relativistic energy and momentum conservation (checked above). Units: h/(mc) has dimension length, so alpha = h/(mc lambda_0) is dimensionless ✓. Eq. 5 follows from Eq. 3: 1/nu_theta = (1 + 2 alpha sin^2)/nu_0 gives lambda_theta - lambda_0 = (2 alpha sin^2) lambda_0 = (2h/mc) sin^2 ✓. Numerical constant: 2h/(mc) = 0.048526 A [calc] vs 0.0484 in Eq. 31 (p.494) and 0.0485 on Fig. 5. The 0.0484 reflects 1923 constants (<0.3% off); abstract's 0.0242 is h/mc, and p.1 uses 0.0242. No inconsistency.

b. Experiment. Compton's p.494 statement that the shift at 90 deg is "roughly 0.03 A" over lambda 0.7 to 0.025 A is a coarse test of independence of lambda_0 (the data are in an unpublished Bull. N.R.C. 20 report, not read). Modern precision measurements confirm the formula [memory: high; unverified by web in this step].

c. Tested? By Compton himself here (graphite/Mo, gamma-rays/Table I); independently verified later (Compton-Simon, Bothe-Geiger above, plus Debye's independent theoretical derivation of the same formula in 1923 [memory: medium; not web-verified]).

d. What would change. Nothing now.

Verdict: Established.

Claim 3. Experiment agrees: Mo K-alpha at 90 deg, 0.022 observed vs 0.024 predicted; RaC gamma rays 0.022 to 0.068 A at 135 deg.

a. Mathematics. Arithmetic OK: 0.0484 sin^2(45 deg) = 0.0242 ✓ [calc 0.02426]. Gamma: 0.022 + 0.0485 sin^2(67.5 deg) = 0.0634 predicted, vs observed 0.068 (Table I, p.496).

b. Experiment. The Mo line comparison is a single spectrum read off a chart (Fig. 4, p.495): scattered peak at 0.730, primary at 0.708, a 0.022 A difference measured by calcite glancing-angle position (peak at about 7 deg). Fig. 4 intensities are plotted "arbitrary units" with a visibly broad scattered peak, so the 0.022 vs 0.024 agreement is about 10% with no stated uncertainty. It is good for a first result. The gamma-ray row is less direct: see Correction 4. The 90 deg row (0.043 observed vs 0.047 calculated) and 135 deg row (0.068 vs 0.063) show scatter of about 8-9% either direction; p.496 calls this "satisfactory accord". The paper gives no error bars.

c. Tested / replicated. Compton's own data; replicated and refined by others [web: https://arxiv.org/pdf/1106.1365 and other sources on the 1925 experiments confirm the effect; precise replication of the 1923 Mo/graphite number not verified]. [memory: high] Later spectrometer work (Compton 1923b, Ross and others) clearly resolved modified and unmodified lines; not web-checked.

d. What would change. Nothing now.

Verdict: Established (the formula); the specific 1923 gamma-ray numbers are crude and partly circular, historically adequate, not independently probative.

Claim 4. Classical Thomson scattering fails for hard X-rays/gamma-rays; true scattering too high by (1+2alpha)^2/(1+alpha); quantum coefficients match Ishino and Hewlett.

a. Mathematics. The algebra of Eqs. 28-30 follows from Eqs. 24-27 and I re-verified: integrating Eq. 27 over the sphere gives total/classical = (1+alpha)/(1+2alpha)^2 numerically at alpha = 0.01, 0.0347, 0.5, 1.09 (0.9708, 0.9048, 0.375, 0.2067 vs formula identical to 4 digits) [calc]. Normalisation of the probability density in Eq. 22: integral = 1.000000 [calc]; k = 3/8 ✓ (integral of (1+cos^2)sin = 8/3). Eq. 24 is consistent with Eq. 23 only if R^2 replaces R in the printed denominator (see Claim 5). Ishino numbers reproduce: with alpha = 1.09, 0.193 sigma_s ratio gives 0.0399 (paper: 0.040), 0.187 gives 0.0386 (0.038), sigma_a gives 0.0208 and 0.0202 (paper 0.021, 0.020) [calc] ✓. Note the ratio sigma_0/sigma_s = (1+2alpha)^2/(1+alpha) = 4.84 at alpha = 1.09. A conceptual caveat. The step "n quanta per second removed" (p.493, between Eqs. 26 and 28) takes the forward intensity from the classical result (Eq. 26) and equates it with the quantum expression Eq. 24. This fixes the normalisation of the whole distribution; it is a modelling choice, not derived from the quantum dynamics. The Klein-Nishina formula, which does derive it, gives a total number cross section of 0.415 sigma_0 at alpha = 1.09 and an energy-scattered cross section of 0.267 sigma_0 [calc, KN formula from memory: high], versus Compton's 0.314 sigma_0 energy removed and 0.207 sigma_0 energy scattered. So Compton's formulas are right at small alpha (0.9048 vs 0.9062 energy-scattered at alpha = 0.0347 [calc]) but off by about 25% for energy scattered at gamma-ray energies, and Ishino-sized agreement at that level (~10%) was partly fortuitous or within the crude data. Hewlett's carbon data: the paper itself says agreement below 0.5 A is "perhaps within the experimental error", and above 0.5 A the data lie above the curve (Fig. 6, p.498), attributed to excess scattering [read: page image p.16].

b. Experiment. Thomson's formula was known to hold for soft X-rays, and to over-predict scattering for hard gamma-rays (the Ishino data show 0.045 vs 0.193, a factor of 4.3). Consistent with Compton's formula and later with Klein-Nishina.

c. Tested? Ishino (1917) and Hewlett (1921) data were pre-existing and used post hoc; no pre-registered test. Later independent tests via Klein-Nishina agreement [memory: medium-high; Klein-Nishina derived 1928, published 1929: https://pmc.ncbi.nlm.nih.gov/articles/PMC5709540/ confirms derivation from Dirac equation].

d. What would change. In 1923: classical electrodynamics could not be the full account of scattering at short wavelength. Done.

Verdict: Split. Qualitative statement (Thomson fails at high photon energy; scattering falls with alpha): Established. The specific coefficients of Eqs. 28-30: Historically superseded but instructive (leading-order correct, replaced by Klein-Nishina; the agreement with Ishino is real but within wide error).

Claim 5. Intensity concentrated forward (Eq. 27) via moving-radiator trick; Barkla-Ayers fit imperfectly.

a. Mathematics. This is the weakest part, as Compton says (p.491): "I have not, however, succeeded in showing rigidly that if two methods of scattering result in the same relative wave-lengths ... they also result in the same relative intensity ... Nevertheless, we shall assume." This is a hidden-assumption step flagged by the author, and it is a non sequitur: matching the frequency shift (Eqs. 3 and 7) does not imply matching the angular intensity. The scheme then (i) assumes a rest-frame dipole distribution (3/8)(1+cos^2 theta') for an unpolarised beam, (ii) Lorentz-transforms it (Eqs. 11-22), (iii) drops the 1/(1-beta cos theta) time-compression factor because each quantum comes from a different electron (p.492), which is a reasoned choice but not obviously correct.

What I verified: Eqs. 14-19 are standard special relativity (aberration, Doppler, D^4 in Eq. 19; the Wien-Stefan check in Eq. 20 is valid for blackbody radiation) [read: page image p.7-9]. Eq. 22 -> Eq. 25 -> Eq. 27 reduction reproduces numerically at alpha = 0.0347 and 1.09 and at three angles (e.g. alpha = 1.09, theta = 2.5 rad: from Eq. 22 0.03594, from Eq. 25 0.03594) [calc]. So the algebra is right. Typo: Eq. 23 is printed with "16 pi R" in the denominator [read: page image p.10]; dimensionally intensity (energy per area per time) needs R^2, and Eq. 24 (with R^2) follows from Eq. 23 only with R^2. Dimensional check: n h nu_0/R^2 is energy/(time area) ✓; with R alone it is not. Compare to Klein-Nishina (energy-weighted, forward = 1) at alpha = 1.09 [calc]: 40 deg: Eq. 27 0.295, KN 0.414; 90 deg: 0.0697 vs 0.0859; 140 deg: 0.0369 vs 0.0570. So Eq. 27 is right at small alpha (alpha = 0.0347: 0.452 vs 0.452 at 90 deg) and underestimates large-angle intensity by 20-35% at gamma-ray energies. This shows the "not quite rigid" method gives a good approximation but not the correct law.

b. Experiment. Paper's own comparison, Fig. 7 / p.501: iron, ionization chamber, gamma rays from RaC. Text: observed 90 deg intensity 0.074 x classical = 0.037 I_0; Eq. 27 gives 0.070 I_0 at 90 deg [calc]; the plotted points lie below the curve over 60-150 deg (Fig. 7, p.500) by up to a factor 2. "Beautiful agreement ... not a single adjustable constant" (p.501) is thus a qualitative statement: the forward concentration is clear, the numbers are off by factors up to about 2, and the measurement itself is an ionization measurement whose wavelength dependence Compton says is unclear ("It is not clear just how the ionization ... will vary"). Barkla-Ayers: 1.04 observed vs 1.136 from Eq. 27 [calc] (Compton says 1.10), "differs ... by more than their probable experimental error" (p.499), and soft-X-ray data show roughly symmetric scattering. His proposed resolution (ionization proportional to lambda^3, so i_theta = i'_theta, p.500) is an ad hoc but reasonable argument at small alpha.

c. Tested? Compton's own data above; no controlled independent test in the paper. Replaced by Klein-Nishina 1928-29, itself verified by later gamma-ray scattering experiments [web: https://pmc.ncbi.nlm.nih.gov/articles/PMC5709540/ for Klein-Nishina derivation; experimental verification of KN: memory: high, not web-checked here].

d. What would change. Nothing now; Eq. 27 would be replaced by the correct quantum electrodynamic distribution (mathematically: the Dirac equation gives the correct Klein-Nishina formula, not obtainable from the moving-radiator trick).

Verdict: Split. Forward-concentration qualitative claim: Established. The specific Eq. 27 derivation: Historically superseded but instructive (a heuristic with an admitted unproved step; right to leading order in alpha, wrong by 20-35% at alpha near 1; replaced by Klein-Nishina). The Barkla-Ayers comparison: Tested, not supported for Eq. 27 as printed (1.04 observed vs 1.10-1.14 predicted, discrepancy beyond stated error) at the level of that single dataset.

Claim 6. Theory holds only for loosely bound electrons in light elements.

a. Mathematics. Consistent. The paper (p.502) says the derivation "tacitly assumed that there are no forces of constraint". A necessary condition is stated clearly: if the recoil kinetic energy is below the binding energy, "there is no chance for the electron to recoil in the manner we have supposed". Quantitatively, recoil energy ~ alpha-dependent; for Mo K-alpha at 90 deg the shift implies recoil of about 0.0242/0.708 x 17.4 keV = 0.6 keV (alpha approx 0.034, E_photon about 17.5 keV; recoil kinetic energy = h nu_0 - h nu_theta = h nu_0 x 2 alpha sin^2/(1 + 2 alpha sin^2) = 17.5 keV x 0.0333 = about 0.57 keV) [calc estimate], far above carbon valence-electron binding energies (~10 eV) but below the carbon K-shell binding (~0.28 keV: memory, medium). The "light elements" restriction is therefore physically coherent. The speculation that the quantum "may spread itself over a large number of electrons" for bound electrons is vague (not a calculation).

b. Experiment. Consistent with the later observation of both modified and unmodified lines, Compton profiles and coherent (Rayleigh, Bragg) scattering [memory: high].

c. Tested? Not tested in the paper. Modern X-ray scattering literature describes it routinely [memory: high; not web-checked].

d. What would change. Nothing; the bound-electron case is treated in standard quantum mechanics.

Verdict: Established (as a stated limitation); the "quantum spreads over many electrons" remark is Not testable as stated.

Summary table

Files used: source folder: Compton/Compton_1923.pdf (read only); scratch calculations in /tmp/compton_m.