Gaussian primes on a design

Claims · 25 dated claims on Gaussian primes on a design, each with its tag and its witness; newest 2026-10-05.

25 claims
  • Proved In base b on both coordinates of Z[i] with F = {0..b-1}^2 less one vector a, the transform of the level-k set is Prod_(j<k) (D_b(b^j t_1) D_b(b^j t_2) - e(<a, b^j t>)); a sign change of one coordinate of m = a - ((b-1)/2, (b-1)/2) or the swap carries its modulus to the modulus at the reflected or swapped t, so every unshifted grid sum and every supremum or infimum over shifts depends on a only through the orbit of m, C(floor((b+1)/2) + 1, 2) orbits, and the one-digit grid sum at shift 0 is exactly 2 at every vector. Witness: coprime.md, section Gaussian primes on a design, bullet The setup.
  • Proved For a square block R = I_1 x I_2 and d in Z[i], Sum_(z in R, d divides z) e(<z, t>) = N(d)^(-1) Sum_(kappa in (conj d)^(-1) Z[i]/Z[i]) G_(I_1)(t_1 + kappa_1) G_(I_2)(t_2 + kappa_2), so every Type I inner sum over a square block is an average of products of two one-dimensional geometric sums. Witness: coprime.md, section Gaussian primes on a design, bullet The box factors Type I.
  • Proved Every Gaussian design missing one digit vector carries a shifted-grid l^1 certificate in units of b^(2k) with alpha_2 < 1/5 from base 72, alpha_2 < 1/6 from base 267 and alpha_2 < 1/8 from base 5054, through the square chain z - 1 = sum_l (lambda_l^+)^2 z^(-l), each wall failing one base below. Witness: coprime.md, section Gaussian primes on a design, bullet The square chain, computed by lab/py/gaussian-digits verb chain.
  • Verified The square uniform window (A (x) A + 1 1^T)/(b^2 - 1) certifies alpha_2 < 1/5 at every missing vector of every base 43..71 and alpha_2 < 1/6 at every base 107..266, base 42 reading 0.200340 and base 106 reading 0.166814; with the square chain the certificate holds from base 43 against 1/5 and from base 107 against 1/6. Witness: lab/py/gaussian-digits, verbs uwin and chain.
  • Proved The normalised Lebesgue sum of the square kernel is the product of two one-dimensional sums, so the square chain root satisfies z/((2/pi) log b) -> 1 while the Morton chain at base b^2 has (z - 1)^2/(2 (2/pi) log b) -> 1, and the ratio of the two chain exponents tends to 2, reading 1.1863 at b = 10 and 1.4211 at b = 10^6. Witness: coprime.md, section Gaussian primes on a design, bullet The box squares the Lebesgue constant, readings by lab/py/gaussian-digits verb chain.
  • Conjecture The two-digit window reading with sampled cell suprema falls below 1/5 first at base 31 and at every orbit from base 33, over all 4046 orbits of bases 3..44, with orbit spread at most 0.0137 from base 6 and at most 0.0034 from base 21. Witness: lab/py/gaussian-digits, verb census.
  • Refuted min_s Sigma_1(s) = 2 at every orbit of every base 3..32 except five orbits at bases 3 and 5: base 7 missing (3, 3) gives Sigma_1 = 1.997526 at s = (0.019396/7, 0.980604/7), base 3 missing (1, 1) reaches 1.948655, and a local scan of abs(s_i) <= 2.5/b^3 finds a dip at every orbit with a zero coordinate of m at every odd base 3..31, 135 orbits, the dips about b^(-3) from the grid and finer than the scan that printed five. Witness: lab/py/gaussian-digits, verb floor.
  • Conjecture Read through the same two-digit windows, a Gaussian design missing one vector has a larger exponent than its Morton image at base b^2, by 0.046 to 0.106 at all 20 rows of b = 8, 12, 16, 24, 32. Witness: lab/py/gaussian-digits, verb morton.
  • Verified Sum_(z in S_k) Lambda(z) divided by (4/pi) kappa_F #S_k, kappa_F = prod_(pi divides b) N(pi)/(N(pi) - 1) times the share of F coprime to b, reads 0.9991..1.0002 at ten designs of bases 3, 4, 5 and 7 with 2.1 * 10^6 to 1.14 * 10^7 points each. Witness: lab/py/gaussian-digits, verb count.
  • Conjecture An exponent count with no lemma written: Vaughan's identity over Z[i] with linear sums sliced into rows would give the minor-arc exponent 1/8, capped at the balanced split by a Type II inner sum over a sharp region saving only N_in^(-1/4), and 1/6 if the Type II inner variable is smoothed. Witness: coprime.md, section Gaussian primes on a design, bullet The gate over Z[i].
  • Conjecture Granted a minor-arc bound abs(S_P) << y^(2 - 2 delta + o(1)) on square blocks, the region-B classes in two dimensions and an effective prime number theorem for Gaussian primes in squares of side x^(1/2) exp(-c sqrt(log x)) in classes mod divisors of b^infinity, the Gaussian prime sum over a design missing one vector is (4/pi) kappa_F A_F(X) (1 + O(exp(-c sqrt(log X)))) from base 43 at delta = 1/5, 107 at 1/6 and 5054 at 1/8. Witness: coprime.md, section Gaussian primes on a design, bullet What the theorem still asks.
  • Verified Baier's bound (38) for a linear sum over a disc sector of Z[i], got by slicing into rows, saves the square root of its range at a generic frequency and no more, and his Gaussian analogue of Vaughan's p^(-1/4) is N(p)^(-1/24 + eps), the absolute value of p to the power -1/12 + eps, his Theorem 4. Witness: coprime.md, section Gaussian primes on a design, bullet What sums over Z[i] give, read at source.
  • Proved For X >= 3, a box R of Gaussian integers with N(z) <= X on it, xi in C and coprime Gaussian integers q != 0, lambda with abs(xi - lambda/q) <= 4/N(q), Sum_(z in R, z != 0) Lambda(z) e(Re(z xi)) << (X^(7/8) + X N(q)^(-1/2) + X^(11/16) N(q)^(1/2) + X^(1/2) N(q)) (log X)^4 with an absolute effective constant, Lambda read on the ideal (z), by Vaughan's identity over ideals at U = V = X^(1/4), Type I and Type II sums sliced into rows, and the spacing of d xi on squares of side abs(q)/16. Witness: coprime.md, section Gaussian primes on a design, bullets The minor-arc lemma over Z[i], Units and Vaughan's identity, Spacing, Type I and Type II.
  • Proved In the block-sum reduction over Z[i] with blocks of side y inside [0, x)^2, X = 2 x^2 >= 256 and a Dirichlet approximation fixed per residue at M = floor(X^(3/16)/sqrt 2), the residues with N(q) >= X^(1/4) contribute << C_F fill^k y^(2 alpha_2 - 1/4) (x/y)^(7/4) (log X)^4 to a block sum, a power saving in y at bounded x/y exactly when alpha_2 < 1/8. Witness: coprime.md, section Gaussian primes on a design, bullet Region A over Z[i].
  • Proved At a grid point xi = r/y of the block-sum reduction with X = 2 x^2 and x/y bounded, below N(q) = X^(1/4) the bound of the minor-arc lemma saves only N(q)^(1/2) at the first approximation lambda/q, and at height h = abs(r q - lambda y) >= 1 a second approximation has y/(2h) <= abs(q') << y/h, where the bound reads X h/y + X^(11/16) y/h + X^(1/2) y^2/h^2, nontrivial only for h of order X^(1/4) and beyond, the sliced Type II term alone leaving h below about X^(3/16) without saving; this reads the bound, not the sum. Witness: coprime.md, section Gaussian primes on a design, bullet The band the lemma leaves.
  • Verified On the box [1, 2048)^2, X = 8380418, with all 370005 prime-power elements and their Lambda, whose whole mass is 1.8884 times the least value over N(q) of X^(7/8) + X N(q)^(-1/2) + X^(11/16) N(q)^(1/2) + X^(1/2) N(q), the exponential sum stays below 0.7456 times that expression, constant 1 and no logarithm, at 1000 frequencies near sampled lambda/q in ten bands of N(q) from X^0 to X^0.45, 0.1170 at X^0.25 and 0.0153 at X^0.35. Witness: lab/py/gaussian-digits, verb minor.
  • Verified Mitsui bounds the sum of e(S(omega z)) over prime elements of a number field in a region of conjugates, a disc for Q(i), by N^n (log N)^(-sigma), sigma >= 3, off the major arcs, a saving by a power of the logarithm and not a power; read in part II as the scanned journal pages through their text layer, at (4.1), (4.2) and the statement of Theorem 5.1, with part I and the proof unread. Witness: coprime.md, section Gaussian primes on a design, bullet Mitsui's bound is a logarithmic saving.
  • Conjecture At every missing digit vector of every base b >= 3, Sum_(z in S_k) Lambda(z) is asymptotic to (4/pi) kappa_F fill^k as k grows, read at 0.9991..1.0002 at ten designs of bases 3, 4, 5 and 7; no proof is claimed at any base, and the note names the three open inputs a proof from base 5054 would ask. Witness: lab/py/gaussian-digits, verb count.
  • Proved For coprime Gaussian integers q != 0, lambda with abs(xi - lambda/q) <= 4/N(q), every H > 0 and eta > 0, #{h in Z[i] : N(h) <= H, dist(h xi, Z[i]) <= eta} <= 2^18 (H/N(q) + 1)(N(q) eta^2 + 1), dist(w, Z[i]) the Euclidean distance from w to Z[i]; by brute force at H = 10^6 over 184 frequencies with N(q) up to 441650 the count reads at most 18.6834 times the bracket, a reading and not the supremum. Witness: coprime.md, section Gaussian primes on a design, bullet The count lemma, checked by lab/py/gaussian-digits verb hcount.
  • Proved With q, lambda, xi as in the count lemma, U_0, V_0 >= 1 and any alpha_v on N(v) < 2 V_0, beta_u on N(u) < 2 U_0, abs(sum_v sum_u alpha_v beta_u e(Re(v u xi))) <= 2 * 10^4 norm2(alpha) norm2(beta) (U_0 + V_0 + N(q) + U_0 V_0/N(q))^(1/2), by Cauchy-Schwarz with a Gaussian majorant, Poisson summation on Z[i] and the count lemma; with alpha the Cauchy-Schwarz extremal for beta random unimodular or Lambda on the annuli N(v) in [V_0, 2 V_0), N(u) in [U_0, 2 U_0), the ratio to the bracket reads at most 1.4444 at five shapes, a reading below the operator norm. Witness: coprime.md bullet The bilinear bound at Vaughan's strength; lab/py/gaussian-digits verb bilinear.
  • Proved For X >= 3, W smooth on C and supported in X^(1/2)/e <= abs(z) <= X^(1/2) with abs(partial^gamma W) <= X^(-abs(gamma)/2) at abs(gamma) <= 10, and coprime Gaussian integers q != 0, lambda with abs(xi - lambda/q) <= 4/N(q), sum_z Lambda(z) W(z) e(Re(z xi)) << (X^(4/5) + X N(q)^(-1/2) + X^(1/2) N(q)^(1/2)) (log X)^(7/2) with an absolute constant, so X^(4/5) (log X)^(7/2) at X^(2/5) <= N(q) <= X^(3/5): Type I by Poisson summation, Type II by separating W(e n) through abs(z)^(i t_M) (z/abs(z))^l at bounded cost. Witness: coprime.md bullets The minor-arc lemma on a smooth annular weight, Type I by Poisson summation, Type II through the characters.
  • Proved For a square R of side y with y <= abs(z) and N(z) <= X on it and an even Y with 4 X^(1/2) < Y <= 8 X^(1/2), S_R(xi) = Y^(-2) sum_(r mod Y) hat 1_R(r/Y) S_W(xi + conj(r)/Y) with W a sum of << 1 + log(X^(1/2)/y) annular weights equal to 1 on R and Y^(-2) sum_r abs(hat 1_R(r/Y)) <= (2 + log Y)^2, so S_R(xi) << X^(4/5) (log X)^(11/2) (1 + log(X^(1/2)/y)) plus Y^(-2) times the sum of abs(hat 1_R(r/Y)) abs(S_W(xi + conj(r)/Y)) over the shifts whose Dirichlet approximation at M = floor(X^(3/10)/sqrt 2) has norm below X^(2/5); region A at the gate 1/5 is not claimed. Witness: coprime.md, section Gaussian primes on a design, bullet The square through annular sums.
  • Conjecture The Fourier l^1 norm in (log abs(z), arg z) of the square [1, 2]^2 smoothed at scale h, the cost of separating W(v u) through abs(z)^(i t_M) (z/abs(z))^l, grows like h^(-1/2): 6.711 at h = 2^(-4) to 37.321 at 2^(-9), by 1.3923..1.4462 per halving, where a polar box grows logarithmically; at that cost the route through a smoothed square and the characters pays max(y h, (y/h)^(1/2) y^(3/2)) >= y^(5/3) on a block of side y and stops at the gate 1/6. Witness: lab/py/gaussian-digits verb sep.
  • Verified On [1, y)^2 at y = 3^6, 3^7, 3^8 with all prime-power elements, the largest abs(S(xi))/y^(8/5) over 40 frequencies in each band of N(q) near y^0.8..y^1.2 falls in every band from y = 729 to 6561, from 0.2648 to 0.1000 near y^0.8 and from 0.1192 to 0.0432 near y^1.2, the box masses being 17.74 to 42.81 times y^(8/5). Witness: lab/py/gaussian-digits verb gate.
  • Proved The two earlier rows on an exponent count with no lemma written and on the prime sum granted a minor-arc bound, the region-B classes and a prime number theorem in squares are superseded: the sliced minor-arc lemma is written and proved at the gate 1/8, the lemma on a smooth annular weight at 1/5, and from base 5054 a proof still asks the inputs (i) to (iii) of the note, among them Lemma A' at a perturbed point in dimension two for region C1, with x a side length, X = 2 x^2 and the prime number theorem in squares of side x exp(-c sqrt(log x)); no proof is claimed at any base. Witness: coprime.md, section Gaussian primes on a design, bullet What a proof from base 5054 still asks.