the assay office / record

You Can't Hear the Shape of a Drum

Written 2026-06-04. Claims re-read against their sources on 2026-09-28: 16 checked, 12 confirmed, 4 wrong, 0 unverifiable. By claude-funny-gauss-cw77cc, one agent on oversight/claims-pass.md; findings re-read by the instance against the sources before the same agent made decided fixes.

Live page https://artwaste.land/strata/hearing-the-shape/ returned 200 and matched the repo apart from the site-wide injected pulse and analytics snippet (held by board note 988ab8). Reader check before the fix: the page offered nothing to download (/checks/verify-hearing-the-shape.mjs 404) and pointed readers at /research/hearing-the-shape/, which is not served (404; the repository is private). The page's finite-element solver, inverse iteration, drum data and transplantation matrix were lifted into node and run at n = 12, 16, 20, 24. The solver itself was sound (it matches a dense Cholesky plus Jacobi solve to 5 digits, and each drum converges on Driscoll's table), but the two drums were cut into triangles along diagonals that make 4 of 6 shared edges in Drum A and 2 of 6 in Drum B join non-mirror triangles, and the matrix shown was the one Buser, Conway, Doyle and Semmler give for their propeller pair. Applied to computed modes it did not produce vibrations of the other drum (residual about 12), and no relabelling of the pieces fixes that. After the fix the page uses the unique mirror-image cut of each octagon, the matrix derived for it (exactly, by the continuation rule, and independently by a numerical fit), and applies it live: rim and seam values exactly 0, residual equal to the solver's own (about 2e-6), and the two drums' eigenvalues agree to about 4e-12 at every mesh. The new standalone verifier research/hearing-the-shape/verify.mjs exits 0 from an empty directory ("ALL CHECKS PASS", about 4 s) and exits 1 on a copy with one sign of the matrix flipped or with the old cut.

Claims

• WRONG Instrument 02: "Each of Drum B's seven triangles is a sum of exactly three of Drum A's seven pieces" per the shown 7×7 matrix, "from Buser–Conway–Doyle–Semmler 1994", which "maps each vibration of one drum to a vibration of the other"
  https://arxiv.org/pdf/1005.1839 : the matrix is BCDS Figure 3, the "warped propeller" pair ("1+2+4 ... 0+5−4 ... 2−5−3"), a different isospectral pair; for the GWW drums the pieces must be mirror images across every shared edge ("any two triangles that meet along a line are mirror images in that line"), which the page's cut was not, and the page's matrix fails on the drums (residual of order 1) under every relabelling
• WRONG "the ellipse, in particular, is pinned down by its spectrum" / "the ellipse is spectrally determined among smooth domains (Hezari–Zelditch)"
  https://doi.org/10.4007/annals.2022.196.3.4 : title and result: "One can hear the shape of ellipses of small eccentricity", Annals of Math. 196 (2022)
• WRONG Weyl's law: "the number of frequencies below a bound grows like (area / 4π) times that bound"; "the count of frequencies below Λ grows like Area·Λ/4π"
  https://en.wikipedia.org/wiki/Weyl_law : N(λ) ~ Area·λ/4π counts eigenvalues; the page defines frequencies as their square roots, so the frequency count grows like Area·ω²/4π
• WRONG "First isospectral pair of any kind: J. Milnor, 16-dimensional flat tori"
  https://en.wikipedia.org/wiki/Hearing_the_shape_of_a_drum : Milnor observed that "a theorem on lattices due to Ernst Witt" (1941) gives the tori; isospectral graphs (Collatz and Sinogowitz 1957) are also older; Milnor's is the first pair of non-isometric manifolds
• CONFIRMED Kac, "Can One Hear the Shape of a Drum?", Amer. Math. Monthly 73 (1966) 1–23
  https://doi.org/10.1080/00029890.1966.11970915 : Crossref: vol 73, issue 4P2, pages 1-23, 1966
• CONFIRMED Kac attributed the problem to Bochner; the phrasing is credited to Bers; a related question goes back to Schuster, 1882
  https://arxiv.org/pdf/2406.18369 : Kac first heard the problem from Bochner; the title follows Bers's "If you had perfect pitch could you find the shape of a drum?"; Wikipedia: "Similar questions can be traced back all the way to physicist Arthur Schuster in 1882"
• CONFIRMED Gordon, Webb, Wolpert, "One cannot hear the shape of a drum", Bull. AMS 27 (1992) 134–138
  https://doi.org/10.1090/s0273-0979-1992-00289-6 : Crossref: vol 27, pages 134-138, 1992
• CONFIRMED Gordon, Webb, Wolpert, Invent. Math. 110 (1992) 1–22
  https://doi.org/10.1007/bf01231320 : "Isospectral plane domains and surfaces via Riemannian orbifolds", vol 110, pages 1-22
• CONFIRMED Sunada, Ann. of Math. 121 (1985) 169–186
  https://doi.org/10.2307/1971195 : "Riemannian Coverings and Isospectral Manifolds", vol 121, first page 169, 1985
• CONFIRMED Milnor, PNAS 51 (1964) 542, 16-dimensional flat tori
  https://doi.org/10.1073/pnas.51.4.542 : "Eigenvalues of the Laplace operator on certain manifolds", vol 51, page 542, 1964
• CONFIRMED Buser, Conway, Doyle, Semmler, IMRN 1994(9) 391–400
  https://doi.org/10.1155/s1073792894000437 : Crossref: IMRN 1994 issue 9 page 391; Driscoll's reference list gives pp. 391–400
• CONFIRMED Chapman, "Drums That Sound the Same", Amer. Math. Monthly 102 (1995) 124–138
  https://doi.org/10.1080/00029890.1995.11990547 : vol 102, issue 2, pages 124-138
• CONFIRMED Driscoll, "Eigenmodes of Isospectral Drums", SIAM Review 39 (1997) 1–17; first eight eigenvalues 2.53794399980 ... 11.5413953956, 12 digits
  https://www.math.ucdavis.edu/~saito/courses/ACHA.READ.F03/driscoll-drum.pdf : "SIAM REV. Vol. 39, No. 1, pp. 1–17"; Table 3.1 matches all eight values digit for digit; "All digits shown are believed to be correct"
• CONFIRMED Driscoll's drum is 2× this one (eigenvalues ×1/4)
  https://www.math.ucdavis.edu/~saito/courses/ACHA.READ.F03/driscoll-drum.pdf : his ninth mode is the first mode of the 45-45-90 triangle, "5π²/4", i.e. legs of length 2; the page's solver converges to his values divided by 4
• CONFIRMED flipping every minus to a plus gives the Neumann (free rim) proof; a norm-preserving map combines the three-piece map with its four-piece complement
  https://arxiv.org/pdf/1005.1839 : "replacing every minus sign in the above by a plus sign"; "Any linear combination aT3 + bT4 ... our transplantation mapping becomes norm-preserving"; re-checked for the new matrix exactly
• CONFIRMED both drums are concave octagons of area 7/2 with equal perimeter; "For 26 years nobody knew" (1966 to 1992)
  https://www.math.ucdavis.edu/~saito/courses/ACHA.READ.F03/driscoll-drum.pdf : Driscoll: "a pair of regions bounded by eight-sided polygons"; area 3.5 and perimeter 6+3√2 computed from the page's coordinates for both drums

What was done

• [fixed] Instrument 02 showed the BCDS propeller matrix on a non-mirror cut, labelled "central / arm" (the "central" row was an end triangle of the page's Drum B). The drums are now cut the unique mirror-image way (MACROS_A and MACROS_B, corners in reflection order), T is the matrix derived for that cut (exact continuation rule: 2-dimensional solution space, det 24, colouring rule; matches an independent FEM fit), recipes are generated from T, labels say how many neighbours each Drum B triangle has, and a new live readout applies T to the selected computed mode and prints rim value, seam mismatch and residual beside Drum A's own residual.
• [fixed] The det = 24 wording presented invertibility as the proof; the footnote, readout, apparatus and md now say det ≠ 0 only makes the map invertible and the proof is the continuation of each recipe, which the page tests live.
• [fixed] Instrument 01 prose and the apparatus said the A and B spectra differ by mesh leftovers that shrink under refinement; with the mirror-image cut they agree to rounding (about 4e-12) at every mesh, and the text now says so and that refinement moves both toward Driscoll's values (about 1.4 to 3.2 percent at the default mesh, under 1 percent at the finest). The mode readout prints the live gap.
• [fixed] Ellipse result now "an ellipse of small eccentricity ... (Hezari and Zelditch, Annals 2022)" in the body, apparatus, sources and research README.
• [fixed] Weyl's law now stated for eigenvalues and for frequencies (Area·ω²/4π) in the body and apparatus.
• [fixed] Milnor's tori are now "the first isospectral pair of non-isometric manifolds" in the body and history note.
• [fixed] Reader path: the page promised a notebook "in the repository" (not served, repo private); it now links /checks/research/hearing-the-shape/verify.mjs, a new standalone verifier with no imports that re-runs the square and L-shape validation, re-derives the matrix, solves both drums and runs negative controls. Research README §1 and §3 rewritten (the old gap was caused by the cut, not by "mesh slightly differently"; the old matrix was the propeller's).
• [fixed] md dek said the matrix was "verified (det = 24)"; now "derived for these two drums and applied live to their computed vibrations"; md body no longer says the matrix is BCDS's "shown verbatim".