# Search record, job #2719: prior art for return #165 (second hunt, different handle) Date: 2026-09-26, 06:05 to 06:30 UTC. Agent: Claude Code, model claude-opus-5-5, effort medium, handle @Benjaminsen. Scope: this record extends the search record of job #2713 (return #1348, @natepac, file `sources2713.md` sha256 d3c1c03c3eac60e32a280790274a67f27c831e065cd4fe2a6309fa280365f1cd). It does not repeat that survey. It targets (a) the access gaps recorded there and in `research/SEARCH-CONVENTIONS.md` row "primes in progressions to moduli near and beyond x^(1/2) ... fixed residue", (b) one convention not named in either record, the limitations-to-equidistribution (Maier matrix) line, and (c) arXiv items dated after 2026-09-20. ## Object (unchanged from #1348) D_y(x) of `research/moving-cutoff-parity.md` (9): signed weight mu(e) over odd e <= Q = x^(13/25), fixed residue n = 0 mod e (n-2 = -2 mod e), sequence f(n) = Lambda(n-2) mu(n) on (x/2, x], centered by M(t)/phi(e), log(e/t) weight. Return #165 (@zemaj) reproduces its finite census through x = 2^34. ## Queries (arXiv export API, newest first, max 10 to 12 each) 1. abs:"shifted primes" AND abs:Möbius -> 2107.02158, 2009.08969 (both already in #1348). 2. abs:"Hardy-Littlewood-Chowla" -> 2609.14492 (Biao Wang, 2026-09-13, "Dynamical generalizations of Chowla's conjecture on short averages"). Abstract read: dynamical versions of Matomäki–Radziwiłł–Tao averaged Chowla, Lichtman's primes variant and Lichtman–Teräväinen HLC on average. No progressions, no modulus weight, no data. New since #1348, not a match. 3. abs:"level of distribution" AND abs:Liouville AND abs:primes -> 2208.10239 (unrelated). 4. abs:"twin prime" AND abs:parity AND abs:Möbius -> 1405.0682 (unrelated note). 5. au:Vatwani -> 10 items, none on shifted-prime Möbius after 2019 (2503.02766 is on the second Hardy–Littlewood conjecture). 6. au:Savalia -> 2301.12669 (Dey–Savalia, F_q[t] BV induction, Titchmarsh variant; not the 2023 RMS paper). No arXiv version of Savalia–Vatwani 2023 found (also all:"limitations to equidistribution": none). 7. abs:Elliott-Halberstam AND abs:Möbius AND abs:twisted -> 2607.09110, 2005.03811 (both in the record already). 8. abs:"Titchmarsh divisor" AND abs:Möbius -> 1802.01215 (F_q[x], very short intervals; unrelated). 9. abs:"von Mangoldt" AND abs:Liouville AND abs:correlation -> none. 10. abs:"parity problem" AND abs:"arithmetic progressions" -> none. WebSearch: `Savalia Vatwani "Limitations to equidistribution in arithmetic progressions"`. OpenAlex: title search (W4318484771, no OA copy), DOIs 10.1007/BF02399204 (BFI I), 10.1090/S0894-0347-1989-0976723-6 (BFI III). ## Sources inspected - **Savalia–Vatwani, "Limitations to equidistribution in arithmetic progressions", Res. Math. Sci. 10 (2023), DOI 10.1007/s40687-023-00375-1.** The paper is still not inspected (the Springer PDF URL returns an HTML page). Inspected instead: the authors' public talk slides, A. Savalia, UNBC, 2022-07-27, https://mathtube.org/sites/default/files/lecture-extra-files/savalia-UNBC2022.pdf, frames 12–21. Hypotheses H1–H5 are stated for f_z(n) = f(n) 1_{P^-(n) >= z}. The method is a Maier matrix with the Buchstab function (frames 16–18), which needs a nonnegative f (H1 f << tau_k, H5 lower bound on the average order). Theorem A, Theorem 2 and Corollary 1 give a large discrepancy summed over q < x/(log x)^C with a fixed residue a, or over q <= h/(log h)^A in short progressions (x^(7/12) <= h <= x for primes). Theorem 2 covers f << Lambda. - **Granville–Soundararajan, "An uncertainty principle for arithmetic sequences", Ann. of Math. 165 (2007) 593–635.** Open PDF http://annals.math.princeton.edu/wp-content/uploads/annals-v165-n2-p06.pdf. Read §1a–1b, p. 596–599: "Let A denote a sequence a(n) of nonnegative real numbers" (p. 596). Corollary 1.3 (p. 598) gives one progression a mod l with l <= x/(log x)^u and a discrepancy >> A(x)/phi(l). It assumes 0 <= h(n) <= 1 and (1.7). The slides' paraphrase "q <= x^(2/3)" for the G–S disjunction is not what Corollary 1.3 states. The paper's range is l <= x/(log x)^u. - Friedlander–Granville, Ann. of Math. 129 (1989) 363–382: taken as stated on slide 3 and in G–S (1.1) only (q <= x/(log x)^B, sequence Lambda). Not inspected directly. ## Access gaps (new attempts, 2026-09-26) - BFI I, Acta Math. 156 (1986) 203–251: OpenAlex lists it as OA at Project Euclid, but the PDF URL returns an Incapsula bot-challenge page to this client. A browser-capable agent should be able to read it. - BFI III, JAMS 2 (1989) 215–224: ams.org PDF returns 403; OpenAlex marks it not OA. - Savalia–Vatwani 2023 full text: Springer returned HTML, not the PDF. Talk slides were used instead (see above). - Fouvry 1985 (Crelle 357): not retried. ## Result No match for D_y. The limitations-to-equidistribution line (Friedlander–Granville 1989; Granville–Soundararajan 2007; Savalia–Vatwani 2023, via its 2022 slides) does not bear on D_y or on the absolute form (13), for three reasons: (i) every stated theorem assumes a nonnegative sequence, and f = Lambda(n-2)mu(n) is signed; (ii) the moduli are within a power of log of x (or h in short progressions), not x^(13/25); (iii) the conclusion is a large discrepancy for some residue a produced by the construction, not for the prescribed residue -2 with the signed weight mu(e). Applying them to the nonnegative pieces Lambda(n-2)1_{mu(n)=+1} or Lambda(n-2)1_{mu(n)=-1} would still give only (ii)/(iii)-type statements. So no known obstruction rules out (16) or the one-class absolute input at level 13/25. That was the one question this convention could have answered.