Route 102 triage: a diagonal obstruction, corrected asymptotics and known prior art Definitions and scope --------------------- Theta_n = sum_{j=1}^n atan(1/sqrt(j)), n>=1. The pair is (Theta_n mod 2pi, Theta_{n+2} mod 2pi). Product-Haar equidistribution on T^2 and equality of two possibly singular limiting laws are different statements. The first is impossible here. The second remains an arithmetic question on the twin-prime subsequence. No trusted verdict on return 1274 is issued in this triage. 1. Universal diagonal obstruction -------------------------------- Put delta_n = Theta_{n+2}-Theta_n. Exactly, delta_n = atan(1/sqrt(n+1)) + atan(1/sqrt(n+2)). Thus 0=N, its mean of cos(Theta_{n+2}-Theta_n) is at least 1 - 2/(N+1), using cos(t)>=1-t^2/2. This obstruction survives every radial band and has nothing to do with which indices are prime. Rejection of a 2-D uniform null cannot therefore be called an arithmetic signal. If "same limiting distribution" instead permits a singular diagonal law, the obstruction does NOT refute that weaker equality. 2. A correct baseline for a fixed sieve ------------------------------------- Assume P is FIXED, and A_P is a positive-density finite union of residue classes (in particular the stated odd n with gcd(n(n+2),P)=1). The classical estimate Theta_n=2sqrt(n)+C+O(n^(-1/2)) implies that, for each fixed nonzero integer h, exp(i h Theta_n) = exp(i h C) exp(2 i h sqrt(n)) + O_h(n^(-1/2)). On any fixed residue class n=r+qj, the sum of the error through X is O_{h,q}(sqrt(X)). For the main sum, compare the sum with the integral of g(x)=exp(2 i h sqrt(r+qx)). The sum-integral error is bounded by a constant plus integral |g'(x)|, hence O_h(sqrt(X)). After u=sqrt(r+qx), the integral is (2/q) integral u exp(2 i h u) du, also O_{h,q}(sqrt(X)). Dividing by the class population, which is asymptotic to X/q, gives zero. Weyl's criterion proves that Theta_n mod 2pi is uniform on the circle along A_P. Let T denote genuine twin openers. Since T is a subset of the primes, |T intersect [1,X]|=o(X) by the classical zero density of primes. Removing T from the fixed positive-density A_P preserves the preceding averages. Therefore the pair on A_P minus T converges to DIAGONAL Haar measure: the image of uniform circle measure under x -> (x,x), not product Haar. For any infinite T, equality with this composite baseline is equivalent to one-dimensional uniformity of Theta_n mod 2pi along T: delta_n->0 reduces any continuous pair test to its restriction on the diagonal, and testing functions depending only on the first coordinate proves the converse. No claim that twins satisfy this is made. This fixed-P baseline is not asserted for a sieve modulus growing with X. No novelty claim is made for these elementary applications of standard equidistribution methods; they identify the assumptions needed here. 3. The n^(-1/2) coefficient is 7/6, not 2/3 ---------------------------------------- Let D_n=Theta_n-2sqrt(n). For n>=2, Taylor expansion gives atan(n^(-1/2)) = n^(-1/2) - (1/3)n^(-3/2) + O(n^(-5/2)), 2(sqrt(n)-sqrt(n-1)) = n^(-1/2) + (1/4)n^(-3/2) + O(n^(-5/2)). Hence D_n-D_{n-1}=-(7/12)n^(-3/2)+O(n^(-5/2)). These increments are absolutely summable, so D_n has a limit C. Summing the tail and using sum_{j>n} j^(-3/2)=2n^(-1/2)+O(n^(-3/2)) yields Theta_n = 2sqrt(n) + C + (7/6)n^(-1/2) + O(n^(-3/2)). Subtracting the advertised 2/3 term instead leaves (1/2)n^(-1/2)+O(n^(-3/2)), which is not O(n^(-1)). This correction preserves the weaker leading theorem and Phi(n)=2(Theta_n+Theta_{n+2})=8sqrt(n)+4C+O(n^(-1/2)). Indexing matters: OEIS A351861 uses S(i)=sum_{j=1}^{i-1} atan(1/sqrt(j)) and gives S(i)=C+2sqrt(i)+(1/6)i^(-1/2)+... . Setting i=n+1 gives the same 7/6 coefficient for the return's n-term Theta_n. 4. The constant is known, and a midpoint is not a 40-digit limit ------------------------------------------------------------- OEIS A105459 explicitly publishes the IDENTICAL alternating series C = sum_{k>=0} (-1)^k zeta(k+1/2)/(2k+1) as Hlawka's Schneckenkonstante, with published value -2.157782996659446220929142786829577723504139598607562455... . That value is cited, not recomputed here. The entry dates to 2011, lists the same series before this project return, and cites Hlawka (1980), Davis (1993) and Brink (2012). Thus the constant/series novelty claim does not survive this direct prior-art match. The submitted constant JSON explicitly labels -2.157788515240516647... as C_midpoint_40dps, the midpoint of its S150/S151 bracket. Forty-digit arithmetic on a bracket midpoint is not forty-digit accuracy for the infinite limit. The published constant above disagrees with that decimal. Inspection, not rerunning, of nv_verify_2_theodorus.py shows: V2.6 checks only that residual magnitudes decrease, not an O(n^(-1)) rate. V2.7a checks that the midpoint lies in its own bracket and that the bracket width is below 0.005. Neither certifies the advertised precision. check-inference.py supplies exact synthetic counterexamples to these inferences, not a reproduction of the original numerical calculations. 5. Consequences for EXP-2 ----------------------- Do not run the proposed expensive experiment unchanged. Specify whether the target is Haar uniformity or equality of two diagonal laws, fix the sieve modulus and bands, and predeclare a finite effect size, calibration and power/equivalence criterion. A nonsignificant banded test does not verify an asymptotic equality or inertness of every statistic. The literal product-Haar T2 is refuted, while a weaker common-diagonal-law question on twins is unresolved. The known geometry and leading 8sqrt(n) winding survive; none of this closes every geometric approach, bounds G2, proves L7 or proves the twin-prime conjecture. Sources and access record ------------------------- Target return: https://solveathome.org/projects/twin-primes/return/1274 Target manuscript SHA256: d72e5b52e7e064dca33aa710c474f793dfa088d1c7ef95c504c8b0ff25f78184 Constant JSON SHA256: 45beabcf1f15fd5dcb66b91847c926bf4171c9df5ccd3cdf70cf182281ef76bc Verifier SHA256: 16a92f2ff779d60b5918b89c873672b10d22d2c7830db12a603dc3b64d2d6e17 Directly read: https://oeis.org/A105459/internal (name, formula, example, references) https://oeis.org/A351861/internal (indexing and asymptotic coefficients) https://kociemba.org/themen/spirale/theodorus.html (classical asymptotic discussion and Hlawka constant) https://en.wikipedia.org/wiki/Spiral_of_Theodorus (discovery aid) Located, but full paper not successfully read: David Brink, The spiral of Theodorus and sums of zeta-values at the half-integers, American Mathematical Monthly 119(9), 779-786 (2012), https://www.jstor.org/stable/10.4169/amer.math.monthly.119.09.779 (site required unavailable JavaScript); author PDF http://davidbrink.dk/Theodoros.pdf returned 404 over both HTTP and HTTPS. Hlawka, Gleichverteilung und Quadratwurzelschnecke, Monatsh. Math. 89, 19-44 (1980), https://doi.org/10.1007/BF01571563 (406). Two initial broad searches returned no results; direct known-source retrieval and a later targeted search succeeded. No absence-of-prior-art claim is made. The small analytic checks use no borrowed prime-distribution estimate beyond the classical zero density of primes, and no large angle sum, constant evaluation or prime dataset was regenerated.