=== PART A: the mixed CRT class is a Kloosterman fraction === c = 0 (n0), 2 (n+), -2 (n-) <=> c/n = 2( inv(n0 n-, n+)/n+ - inv(n0 n+, n-)/n- ) mod 1 <=> c = 2 n0 v, v = inv(n0) (n+), v = -inv(n0) (n-), so v^2 = inv(n0)^2 (n+ n-) random coprime triples tested: 3000 worst residual (fraction form) = 0 (root form) = 0 => with n- = 1 the class is c/n = 2 inv(n0)/n+ mod 1, the phase of the additive divisor problem. === PART B: the class mean of R_n is zero === R_n(c) = [(r-c)^+ + (r+c-n)^+ - r^2/n]/L, r = L mod n; sum_{c mod n} R_n(c) = 0 random (L,n) tested: 2000 worst |class sum| = 1.87e-12 => MAIN = L w/n is exactly the average of COUNT over the n classes: REM is a discrepancy, not a size. === PART C: X by branch type and modulus band, five exact levels === x = 7 L = 210 y = 13 ln y = 2.565 primes 7..y: 3 descent nodes = 64 type | band | COUNT | MAIN | REM | delta*REM n=1 | n<=2L | 162.145455 | 162.145455 | 0.000000 | 0.000000 n=1 | 2LL ln^2 y | 0.000000 | 0.000000 | 0.000000 | 0.000000 n=1 | ALL | | | 0.000000 | 0.000000 c0 | n<=2L | 28.913401 | 23.162779 | 5.750621 | 0.189581 c0 | 2LL ln^2 y | -0.000000 | 0.000000 | -0.000000 | -0.000000 c0 | ALL | | | 6.234001 | 0.205517 pure | n<=2L | 22.235690 | 22.794245 | -0.558555 | -0.018414 pure | 2LL ln^2 y | 0.000000 | -0.000000 | 0.000000 | 0.000000 pure | ALL | | | -0.560269 | -0.018470 mix2a | n<=2L | 0.315152 | 1.474135 | -1.158984 | -0.038208 mix2a | 2LL ln^2 y | 0.000000 | 0.000000 | -0.000000 | -0.000000 mix2a | ALL | | | -1.158323 | -0.038186 mix2b | n<=2L | 0.481481 | 0.368534 | 0.112948 | 0.003724 mix2b | 2LL ln^2 y | 0.000000 | 0.000000 | -0.000000 | -0.000000 mix2b | ALL | | | 0.107805 | 0.003554 mix3 | n<=2L | 0.000000 | 0.000000 | 0.000000 | 0.000000 mix3 | 2LL ln^2 y | 0.000000 | -0.000000 | 0.000000 | 0.000000 mix3 | ALL | | | -0.010285 | -0.000339 sum MAIN = 210.000000 (= L = 210) sum REM = 4.612929 corpus X = 4.612929 rel diff = 6.4e-8 theta2-step groups: delta*X1 = 0.205517 (corpus 0.205517) delta*(X - X1) = -0.053442 (corpus 2*dX2 + dXmix = -0.053442) symmetric split of X - X1: delta*X2 each = -0.009235 (corpus column 0.094199) delta*Xmix = -0.034971 (corpus column -0.24184) [see PART F] x = 11 L = 2310 y = 47 ln y = 3.850 primes 7..y: 12 descent nodes = 6544 type | band | COUNT | MAIN | REM | delta*REM n=1 | n<=2L | 1677.179717 | 1677.179717 | 0.000000 | 0.000000 n=1 | 2LL ln^2 y | 0.000000 | 0.000000 | 0.000000 | 0.000000 n=1 | ALL | | | 0.000000 | 0.000000 c0 | n<=2L | 311.477549 | 299.644005 | 11.833544 | 0.201188 c0 | 2LL ln^2 y | 2.245429 | 0.003563 | 2.241867 | 0.038115 c0 | ALL | | | 17.270088 | 0.293618 pure | n<=2L | 292.200557 | 291.655599 | 0.544958 | 0.009265 pure | 2LL ln^2 y | 0.000000 | 0.000475 | -0.000475 | -0.000008 pure | ALL | | | 0.530367 | 0.009017 mix2a | n<=2L | 30.439937 | 32.473780 | -2.033843 | -0.034578 mix2a | 2LL ln^2 y | 0.001364 | 0.027856 | -0.026493 | -0.000450 mix2a | ALL | | | -2.224489 | -0.037820 mix2b | n<=2L | 7.784039 | 7.988406 | -0.204367 | -0.003475 mix2b | 2LL ln^2 y | 0.002867 | 0.003088 | -0.000221 | -0.000004 mix2b | ALL | | | -0.210036 | -0.003571 mix3 | n<=2L | 0.260086 | 0.520157 | -0.260071 | -0.004422 mix3 | 2LL ln^2 y | 0.009585 | 0.020771 | -0.011185 | -0.000190 mix3 | ALL | | | -0.292742 | -0.004977 sum MAIN = 2310.000000 (= L = 2310) sum REM = 15.073187 corpus X = 15.073187 rel diff = 1.0e-8 theta2-step groups: delta*X1 = 0.293618 (corpus 0.293618) delta*(X - X1) = -0.037351 (corpus 2*dX2 + dXmix = -0.037351) symmetric split of X - X1: delta*X2 each = 0.004509 (corpus column 0.047843) delta*Xmix = -0.046368 (corpus column -0.133037) [see PART F] x = 13 L = 30030 y = 173 ln y = 5.153 primes 7..y: 37 descent nodes = 381028 type | band | COUNT | MAIN | REM | delta*REM n=1 | n<=2L | 21532.593050 | 21532.593050 | 0.000000 | 0.000000 n=1 | 2LL ln^2 y | 0.000000 | 0.000000 | 0.000000 | 0.000000 n=1 | ALL | | | 0.000000 | 0.000000 c0 | n<=2L | 4023.453685 | 4006.769716 | 16.683969 | 0.169052 c0 | 2LL ln^2 y | 7.059861 | 0.004393 | 7.055469 | 0.071490 c0 | ALL | | | 31.799107 | 0.322208 pure | n<=2L | 3891.395196 | 3891.104550 | 0.290646 | 0.002945 pure | 2LL ln^2 y | 0.000000 | 0.000521 | -0.000521 | -0.000005 pure | ALL | | | 0.269426 | 0.002730 mix2a | n<=2L | 470.304366 | 472.531600 | -2.227234 | -0.022568 mix2a | 2LL ln^2 y | 0.008549 | 0.039080 | -0.030531 | -0.000309 mix2a | ALL | | | -2.592985 | -0.026274 mix2b | n<=2L | 115.733337 | 115.665166 | 0.068170 | 0.000691 mix2b | 2LL ln^2 y | 0.002466 | 0.003872 | -0.001406 | -0.000014 mix2b | ALL | | | 0.064437 | 0.000653 mix3 | n<=2L | 9.994723 | 9.953906 | 0.040818 | 0.000414 mix3 | 2LL ln^2 y | 0.026312 | 0.035466 | -0.009155 | -0.000093 mix3 | ALL | | | 0.017943 | 0.000182 sum MAIN = 30030.000000 (= L = 30030) sum REM = 29.557928 corpus X = 29.557928 rel diff = 1.6e-9 theta2-step groups: delta*X1 = 0.322208 (corpus 0.322208) delta*(X - X1) = -0.022709 (corpus 2*dX2 + dXmix = -0.022710) symmetric split of X - X1: delta*X2 each = 0.001365 (corpus column 0.027019) delta*Xmix = -0.025439 (corpus column -0.076748) [see PART F] x = 17 L = 510510 y = 709 ln y = 6.564 primes 7..y: 124 descent nodes = 20716231 type | band | COUNT | MAIN | REM | delta*REM n=1 | n<=2L | 364970.594249 | 364970.594249 | 0.000000 | 0.000000 n=1 | 2LL ln^2 y | 0.000000 | 0.000000 | 0.000000 | 0.000000 n=1 | ALL | | | 0.000000 | 0.000000 c0 | n<=2L | 68579.611090 | 68557.302103 | 22.308988 | 0.141827 c0 | 2LL ln^2 y | 17.067830 | 0.004928 | 17.062902 | 0.108476 c0 | ALL | | | 54.400025 | 0.345842 pure | n<=2L | 66543.623498 | 66543.363492 | 0.260006 | 0.001653 pure | 2LL ln^2 y | 0.000000 | 0.000502 | -0.000502 | -0.000003 pure | ALL | | | 0.231407 | 0.001471 mix2a | n<=2L | 8234.840617 | 8237.444645 | -2.604028 | -0.016555 mix2a | 2LL ln^2 y | 0.017092 | 0.051062 | -0.033970 | -0.000216 mix2a | ALL | | | -3.116682 | -0.019814 mix2b | n<=2L | 2013.933685 | 2013.938611 | -0.004926 | -0.000031 mix2b | 2LL ln^2 y | 0.002992 | 0.004426 | -0.001434 | -0.000009 mix2b | ALL | | | -0.009842 | -0.000063 mix3 | n<=2L | 184.590365 | 184.649495 | -0.059130 | -0.000376 mix3 | 2LL ln^2 y | 0.045562 | 0.056189 | -0.010628 | -0.000068 mix3 | ALL | | | -0.104652 | -0.000665 sum MAIN = 510510.000000 (= L = 510510) sum REM = 51.400255 corpus X = 51.400255 rel diff = 8.0e-9 theta2-step groups: delta*X1 = 0.345842 (corpus 0.345842) delta*(X - X1) = -0.019071 (corpus 2*dX2 + dXmix = -0.019071) symmetric split of X - X1: delta*X2 each = 0.000736 (corpus column 0.017012) delta*Xmix = -0.020542 (corpus column -0.053095) [see PART F] x = 19 L = 9699690 y = 3109 ln y = 8.042 primes 7..y: 440 descent nodes = 135567709 type | band | COUNT | MAIN | REM | delta*REM n=1 | n<=2L | 6930299.026552 | 6930299.026552 | 0.000000 | 0.000000 n=1 | 2L8L | 0.000000 | 0.000000 | 0.000000 | 0.000000 n=1 | ALL | | | 0.000000 | 0.000000 c0 | n<=2L | 1304301.785360 | 1304273.635893 | 28.149467 | 0.120266 c0 | 2L8L | 46.403983 | 0.053303 | 46.350680 | 0.198030 c0 | ALL | | | 85.044930 | 0.363348 pure | n<=2L | 1265828.833148 | 1265828.614032 | 0.219116 | 0.000936 pure | 2L8L | 0.000000 | 0.006188 | -0.006188 | -0.000026 pure | ALL | | | 0.184578 | 0.000789 mix2a | n<=2L | 157275.293600 | 157278.567353 | -3.273753 | -0.013987 mix2a | 2L8L | 0.294673 | 0.517526 | -0.222853 | -0.000952 mix2a | ALL | | | -3.936989 | -0.016820 mix2b | n<=2L | 38445.026261 | 38445.021860 | 0.004401 | 0.000019 mix2b | 2L8L | 0.040553 | 0.047115 | -0.006562 | -0.000028 mix2b | ALL | | | -0.003166 | -0.000014 mix3 | n<=2L | 3560.597026 | 3560.549998 | 0.047028 | 0.000201 mix3 | 2L8L | 0.515313 | 0.547456 | -0.032143 | -0.000137 mix3 | ALL | | | -0.000218 | -0.000001 sum MAIN = 9699690.000000 (= L = 9699690) sum REM = 81.289135 corpus X = 81.289135 rel diff = 1.3e-11 theta2-step groups: delta*X1 = 0.363348 (corpus 0.363348) delta*(X - X1) = -0.016046 (corpus 2*dX2 + dXmix = -0.016046) symmetric split of X - X1: delta*X2 each = 0.000394 (corpus column 0.011379) delta*Xmix = -0.016835 (corpus column -0.038804) [see PART F] === PART D: the n > 2L COUNT of the mixed types, rebuilt from the ROOTS === COUNT_{>2L}(type) = sum_{(n,eps) in type, n > 2L} sum_{c5} 6 f5(c5) D_y lam.lam.lam (L - ||c'||_{30n})^+ / L, c' = class mod 30n x = 7 (n,eps) with n > 2L in the mixed types: 24 mix2a from the roots = 0.031515 from the h-sum = 0.031515 diff = 6.9e-18 mix2b from the roots = 0.000000 from the h-sum = 0.000000 diff = 0.0e+0 mix3 from the roots = 0.000000 from the h-sum = 0.000000 diff = 0.0e+0 x = 11 (n,eps) with n > 2L in the mixed types: 16763694 mix2a from the roots = 0.106093 from the h-sum = 0.106093 diff = 1.4e-17 mix2b from the roots = 0.041054 from the h-sum = 0.041054 diff = -2.1e-17 mix3 from the roots = 0.086297 from the h-sum = 0.086297 diff = -4.2e-17 => above 2L the mixed COUNT is the mass of the root measure inside the window bump, exactly. === PART E: sizes against ln y === mixed = mix2a + mix2b + mix3. REM by band; MAIN of mix3 and of mix2a+mix2b above 2L x | ln y | REM mix n<=2L | REM mix n>2L | REM mix total | /ln y | /ln^2 y | MAIN mix3 >2L | /ln^3 y | MAIN mix2 >2L | /ln^2 y 7 | 2.565 | -1.0460 | -0.0148 | -1.0608 | -0.4136 | -0.1612 | 0.0103 | 0.0006 | 0.0360 | 0.0055 11 | 3.850 | -2.4983 | -0.2290 | -2.7273 | -0.7084 | -0.1840 | 0.1190 | 0.0021 | 0.3435 | 0.0232 13 | 5.153 | -2.1182 | -0.3924 | -2.5106 | -0.4872 | -0.0945 | 0.4391 | 0.0032 | 0.8081 | 0.0304 17 | 6.564 | -2.6681 | -0.5631 | -3.2312 | -0.4923 | -0.0750 | 1.0141 | 0.0036 | 1.4843 | 0.0345 19 | 8.042 | -3.2223 | -0.7180 | -3.9404 | -0.4900 | -0.0609 | 1.9558 | 0.0038 | 2.3436 | 0.0362 the absolute size of the mixed REM by type, and the two bands against ln y: x | mix2a REM | mix2b REM | mix3 REM | mix n<=2L /ln y | mix n>2L /ln y | 2X2 | X1 | delta*Xmix 7 | -1.1583 | 0.1078 | -0.0103 | -0.4078 | -0.0058 | -0.5603 | 6.2340 | -0.034971 11 | -2.2245 | -0.2100 | -0.2927 | -0.6489 | -0.0595 | 0.5304 | 17.2701 | -0.046368 13 | -2.5930 | 0.0644 | 0.0179 | -0.4110 | -0.0761 | 0.2694 | 31.7991 | -0.025439 17 | -3.1167 | -0.0098 | -0.1047 | -0.4065 | -0.0858 | 0.2314 | 54.4000 | -0.020542 19 | -3.9370 | -0.0032 | -0.0002 | -0.4007 | -0.0893 | 0.1846 | 85.0449 | -0.016835 the mixed REM by band as a share of the mixed REM total, and the mixed MAIN above 2L by type: x = 7 n<=2L: 0.9861 2LL ln^2 y: 0.0000 | MAIN>2L mix2a 0.0309 mix2b 0.0051 mix3 0.0103 pure 0.0017 c0 0.0069 [L ln^2 y = 6.579 L] x = 11 n<=2L: 0.9160 2LL ln^2 y: 0.0139 | MAIN>2L mix2a 0.2967 mix2b 0.0467 mix3 0.1190 pure 0.0146 c0 0.0613 [L ln^2 y = 14.824 L] x = 13 n<=2L: 0.8437 2LL ln^2 y: 0.0164 | MAIN>2L mix2a 0.7158 mix2b 0.0923 mix3 0.4391 pure 0.0212 c0 0.1135 [L ln^2 y = 26.556 L] x = 17 n<=2L: 0.8257 2LL ln^2 y: 0.0142 | MAIN>2L mix2a 1.3324 mix2b 0.1519 mix3 1.0141 pure 0.0286 c0 0.1805 [L ln^2 y = 43.084 L] x = 19 n<=2L: 0.8178 2L8L: 0.0664 | MAIN>2L mix2a 2.1278 mix2b 0.2158 mix3 1.9558 pure 0.0345 c0 0.2504 [L ln^2 y = 64.675 L] === PART F: the varE-theta2-step X2 column, rebuilt, against the symmetric group sum === W_-(h) := 6[6|h] f5(h) D_y prod_{p | h-2} (p-3)/(p-4), W_+(h) the same with h+2; b2 = mean of W_-/6 over 6|h corpus column: X2c = [h=0 term] + 2 sum_{h>0} (1-h/L) (W_-(h) - 6 b2) (positive half of one pattern, doubled) group sum: X2 = [h=0 term] + sum_{h>0} (1-h/L) (W_-(h) + W_+(h) - 12 b2) (= sum over |h| < L of W_- - 6 b2, since W_-(-h) = W_+(h)) x = 7 delta*X2c = 0.094199 (corpus column 0.094199) delta*X2 = -0.009235 (this file, PART C: -0.009235) X2c = 2.8574 X2 = -0.2801 2X2 = -0.5603 x = 11 delta*X2c = 0.047843 (corpus column 0.047843) delta*X2 = 0.004509 (this file, PART C: 0.004509) X2c = 2.8140 X2 = 0.2652 2X2 = 0.5304 x = 13 delta*X2c = 0.027019 (corpus column 0.027019) delta*X2 = 0.001365 (this file, PART C: 0.001365) X2c = 2.6666 X2 = 0.1347 2X2 = 0.2694 x = 17 delta*X2c = 0.017012 (corpus column 0.017012) delta*X2 = 0.000736 (this file, PART C: 0.000736) X2c = 2.6760 X2 = 0.1157 2X2 = 0.2314 x = 19 delta*X2c = 0.011379 (corpus column 0.011379) delta*X2 = 0.000394 (this file, PART C: 0.000394) X2c = 2.6633 X2 = 0.0923 2X2 = 0.1846 => the corpus X2 column doubles the positive half of the h = 2 (p) pattern; the two patterns are mirror images, not even functions of h. X = X1 + 2 X2 + Xmix holds in both bookkeepings; the corpus Xmix absorbs the difference 2(X2c - X2). === PART G: the mix2a cell below 2L, split by balance of the two parts === mix2a = (n0 | h, e | h-+2), n = n0 e <= 2L. UNBALANCED: min(n0, e) <= L^(2/5); BALANCED: min(n0, e) > L^(2/5) (the trilinear Kloosterman-fraction bound of Bettin-Chandee beats the trivial bound only on the balanced side; see the note, section 4) x | L^(2/5) | unbal COUNT | unbal MAIN | unbal REM | bal COUNT | bal MAIN | bal REM | unbal REM / cell REM 7 | 8.490 | 0.2206 | 1.3301 | -1.1095 | 0.0945 | 0.1440 | -0.0494 | 0.9573 11 | 22.154 | 30.0687 | 31.9642 | -1.8956 | 0.3713 | 0.5095 | -0.1383 | 0.9320 13 | 61.805 | 468.5906 | 470.6924 | -2.1018 | 1.7138 | 1.8392 | -0.1254 | 0.9437 17 | 191.956 | 8230.0641 | 8232.6331 | -2.5690 | 4.7765 | 4.8115 | -0.0350 | 0.9865 19 | 623.309 | 157265.2720 | 157268.5370 | -3.2650 | 10.0216 | 10.0303 | -0.0087 | 0.9973