A hyperelliptic curve of genus 3 over Q with at least 212 rational points, whose Jacobian has no endomorphisms other than the integers. As far as we know, the most found before on a hyperelliptic curve of genus 3 was 176: by Keller and Kulesz in 1995, on a curve with sixteen automorphisms, and by Elkies in 2015, on a curve with none but the hyperelliptic involution. The new curve lies on the K3 surface of discriminant −163, as a section of a double cover of P1 × P1. Two searches of such sections are described here. The first, by size of coefficients, found a curve with 210 points. The second, through small rational points of the surface, found this one. All 212 points can be regenerated from the 14 values of x that the model forces, by a rule simple enough to run in your browser.
A curve of genus at least 2 has only finitely many rational points [6], and a conjecture of Bombieri and Lang would imply that the number is bounded in terms of the genus alone [2]. No bound is known, so the largest number found on a single curve is a record people keep: for each genus, and in genus 3 separately for hyperelliptic curves and for plane quartics [4].
There are two kinds of record. Keller and Kulesz found 176 points on y2 = 7920000(x2 + 1)4 − 136782591x2(x2 − 1)2 [7], [5]. That curve has sixteen automorphisms, and its points form 11 orbits of sixteen. Elkies matched the count in 2015 with a curve that has no symmetry beyond y ↦ −y [4]. Its octic has Galois group S8, so by the criterion below its Jacobian has endomorphism ring Z. Such curves are harder to fill with points, because a point brings along only its mirror image.
The curves of this note are of the second kind. We have not found a larger count in the literature for either kind, but our check of the literature is limited to the sources cited here and to web searches, and proves nothing.
Elkies' curve and ours lie on the same surface: the K3 surface X of Picard number 20 and discriminant −163, which is also behind the records for genus 2 [3], [11]. A K3 surface has many projective models. The ones used here are double covers of P1 × P1,
with P of degree 4 in s and of degree 4 in t. A curve of bidegree (1,1) on P1 × P1 is the graph of a Möbius transformation t = μ(s). Substituting it for t and clearing the denominator, which is a fourth power, gives w2 = (an octic in s): a hyperelliptic curve of genus 3. So the model carries a three-parameter family of such curves, one for each μ, the way a quartic surface carries its plane sections. Elkies' search [4] was of this kind.
The model of this note is
and the curve is its section by 31st − 38s + 166t − 168 = 0, that is, t = (38s + 168)/(31s + 166): one finds (31s + 166)4 P(s, t) = f(s), where y2 = f(x) is the equation at the top.
Rational curves on the surface force rational points on every section. Over the vertical line s = s0 the model is w2 = P(s0, t), and for seven values of s0 this quartic in t is the square of a quadratic, so that the surface splits over the line into two rational curves w = ±q(t). The same happens over seven horizontal lines. A section meets each of these 14 split rulings once, and at that point P is a square: 14 values of x, and 28 points, before any search. In the same way the surface splits over certain curves of bidegree (1,1); the catalogue of the model lists 63. A section meets such a curve in two points, the roots of a quadratic, and when the roots are rational they give two more values of x. On our section 14 of the 63 quadratics have rational roots, and they give 26 values of x that the rulings do not. Together the model forces 40 of the 106 values of x.
Elkies used the models of X whose branch curve is smooth, of which he says there are only two [4]. Models with a singular branch curve are far more numerous, and quartic models supply them. Let S ⊂ P3 be a smooth quartic model of X that contains two skew rational lines. The planes through one line and the planes through the other give a map S → P1 × P1 of degree 2. In coordinates in which the lines are X = Y = 0 and Z = T = 0, the quartic is A + B + C with A, B, C of bidegrees (3,1), (2,2), (1,3) in (X, Y; Z, T), and the double cover is
No new lattice computation is needed: a catalogue of the lines and conics on S gives the split rulings and the split (1,1)-curves of the double cover.
An earlier phase of this project had computed such catalogues for quartic models of X [12]. On 1,723 of them we listed 63,881 pairs of skew lines. Every pair whose double cover has a smooth branch curve gave one of Elkies' two models. The others we sorted by the number of rational curves of degree 1 on the double cover, two for each split ruling, and into classes by their numbers of points modulo four primes. There are 434 classes with at least 30 such curves, and 1,800 with 28. The model above is one of the 1,800: its 28 curves of degree 1 are 18 lines and 10 conics of a quartic with 34 lines.
On each of the 434 classes we looked at every section a·st + b·s + c·t + d = 0 with integers of absolute value at most 40: about 20 million sections. Each is scored by its number of forced values of x, which costs a few microseconds. The sections with the best scores are then counted: all points with x of height at most 600 are found with Stoll's ratpoints [9]. Those with at least 130 points are counted again to height 2 · 104, in several coordinate systems, and closed under the five-point rule of a later section. Every curve with at least 150 points is then put through the endomorphism test.
One section stood out at once. On a model with one node and 16 split rulings, the section 12st − 7s + 4t − 9 = 0 is the curve
with 210 rational points: 105 values of x, all of height at most 3563, and none at infinity. Its Jacobian also has endomorphism ring Z.
The same search ran on Elkies' two smooth models, where it recovered his curve, and on models with nodes that come from the sextic double planes of the genus 2 project [11]. Of the 1,800 classes with 28 curves of degree 1 it took the first 600, in order of the size of their coefficients, found at best 166 points, and stopped. The model of this note is number 928 in that order.
The first search looks at a section only if its coefficients are small, and counts its points only if the catalogue forces many of them. The second search drops both conditions. The rational points of the curve are the rational points (s, t) of the surface that lie on the section. So list the rational points of the surface with small s and t, those where P(s, t) is a square, and look for sections that pass through many of them. Two points lie on a one-parameter family of sections, and each further point lies on exactly one member of the family. A member chosen by many points is a candidate, whatever the size of its coefficients.
The model of this note has 192 such points with s and t of height at most 7, away from the split rulings, and so 18,336 pairs. One pair is (s, t) = (−6, 3) and (−5, −2). The section of this note passes through both, and through 14 more of the 22,046 points of height at most 50. Its coefficients, 31, −38, 166 and −168, are far outside the range of the first search. It has 178 points with x of height at most 600, and the recount to height 2 · 104 gave 212.
This search ran on all 2,236 classes: the 434, the 1,800 and Elkies' two. With some variants and further models tried in the same way, it produced 121 new curves with at least 150 points and endomorphism ring Z: the curve of this note, and among the others curves with 206, 194, 186 and 184 points. With the 133 from the first search that makes 254.
Six further sections with 152 to 196 points failed the endomorphism test. Each has an involution other than the hyperelliptic one, so its Jacobian splits, and it does not count here.
Known points produce new ones. Take five rational points with different x. There is one polynomial G of degree at most 4 with y = G(x) at all five, and then
with r of degree at most 3. Each rational root of the cubic r is the x-coordinate of a rational point (x, G(x)). Changing the signs of some of the five y-values changes G, so five values of x give sixteen cubics. A point at infinity can be one of the five: the condition there is that the leading coefficient of G is ±13734, a square root of the leading coefficient of f. In the Jacobian, the three new points are the effective divisor in the class 2K − (P1 + ⋯ + P5), where K is the canonical class. This is the genus 3 form of the four-point rule used for genus 2 in [11].
For example, the points with x = −1, 0, −2, 3/2 and 1/4 and positive y lie on
and
The last three factors give three more points, with x = −16/11, −66/31 and −56/71.
The rule reaches every point we know. Take the 14 values of x on the split rulings. Apply the rule to every five of the 24 known values of smallest height, add what it finds, and repeat. After two rounds the list has all 106 values, and a third round adds nothing. The eight values x = ∞, 0, ±1, ±2, −3 and 2/3 generate the same list.
The rule also finds nothing beyond the list. Applied to every five of the 106 values with every choice of signs, 1,621,454,016 cases, it gives no new point.
The program below runs the five-point rule in your browser, in exact integer arithmetic. It starts from the 14 forced values of x, checks each point it finds by testing that c8f(a/c) is a perfect square, and stops when a round finds nothing new. You can change the starting values and the number of small points it combines. When it has finished, every point remembers the five points that first produced it; pick one to see that relation and follow it back to the starting values.
Hollow: not found yet. Filled: found, darker for earlier rounds. Starting values are ringed.
| statement | status | how |
|---|---|---|
| at least 212 rational points | proved | each of the 105 finite values of x makes f a nonzero square, checked exactly (and again when this page loads). The leading coefficient is 188622756 = 137342, so there are two rational points at infinity |
| the octic has Galois group S8 | proved | it is irreducible modulo 37, and modulo 43 and 79 it factors into irreducibles of degrees 1, 7 and 1, 2, 5. A transitive group with a 7-cycle is doubly transitive, and the fifth power of the last Frobenius element is a transposition |
| End(J) = Z over Q̄ | proved | Zarhin's theorem [10] for y2 = f(x) with Galois group S8. So J is absolutely simple, and the curve has no automorphism over Q̄ except the hyperelliptic involution |
| rank J(Q) ≥ 25 | proved | the differences of the known points generate a subgroup of rank exactly 25. At least 25: 24 quadratic characters of J(Q)/2J(Q), from functions (x − r0)/(x − r1) modulo primes (the test described in [4]), are independent on it; the class P − ιP is killed by all of them and is not in 2J(Q); and J(Q) has no 2-torsion. At most 25: 81 independent five-point relations among the 106 values of x |
| exactly 212 rational points | not proved | a search finds no other point with x of height at most 224, and the five-point rule gives none. Chabauty's method needs the rank to be less than the genus, 3 |
| no hyperelliptic curve of genus 3 over Q is known with more | not proved | we found none in the literature; see the introduction |
For the curve with 210 points the same statements hold with the primes 61, 59 and 37, with rank at least 26, and with no other point found up to height 224. Its 16 forced values of x regenerate all 105 by the five-point rule.
A section through a rational point of the branch curve has a rational Weierstrass point, and can be written y2 = (a polynomial of degree 7). Elkies gave such a curve with 141 known points [4]; a search of his equation to height 30,000 finds 143. The searches of this note found three different curves of this kind with 149 points each, all with Galois group S7 and so with endomorphism ring Z. Their equations are under Data.
The points are quick to find again: a search to height 104 finds all of them. In Magma [1]:
P<x> := PolynomialRing(Rationals());
f := 188622756*x^8 + 1346614668*x^7 - 1919714687*x^6 - 25817308944*x^5 - 13164957080*x^4
+ 104457822336*x^3 + 86305203088*x^2 - 88502034432*x + 19537330176;
C := HyperellipticCurve(f);
pts := Points(C : Bound := 10000);
#pts; // 212
[[Degree(g[1]) : g in Factorization(PolynomialRing(GF(p))!f)] : p in [37, 43, 79]];
// [8], [1, 7], [1, 2, 5]
In PARI/GP [8] (run for this note with version 2.19; it counts the 210 affine points):
f = 188622756*x^8 + 1346614668*x^7 - 1919714687*x^6 - 25817308944*x^5 - 13164957080*x^4 \
+ 104457822336*x^3 + 86305203088*x^2 - 88502034432*x + 19537330176;
print(#hyperellratpoints(f, 10000)) \\ 210, and the 2 points at infinity
print(apply(p -> apply(poldegree, factormod(f, p)[,1]~), [37, 43, 79])) \\ [[8], [1, 7], [1, 2, 5]]
The point counts use Stoll's ratpoints 2.2.1 [9]; the lattice and model computations use PARI/GP and SageMath. The first search used about 1,370 core-hours on cloud machines, and the second search with its follow-up computations about 1,070.
The searches, the computations and this note were produced by a team of AI agents: Lor (Claude Opus 5.5, Anthropic) led the project, with Claude and GPT agents as researchers, engineers and reviewers. The second search was led by a GPT-6.2 Astra agent (OpenAI). The text and figures of this note were written by Claude Fable 5.2 and checked against the project's files. The team worked under the direction of Andrew V. Sutherland, who posed the problem and supervised the work.
One line per x-value, in order of height: (a : b : c) stands for the two points with x = a/c and y = ±b/c4; (1 : 13734 : 0) stands for the two points at infinity.
(−1 : 315375 : 1) (1 : 13734 : 0) (0 : 139776 : 1) (1 : 287109 : 1) (−2 : 273264 : 1) (2 : 290544 : 1) (−3 : 584559 : 1) (2 : 12486000 : 3) (3 : 6727500 : 2) (−4 : 755856 : 1) (1 : 16905750 : 4) (4 : 2118000 : 1) (−5 : 703131 : 1) (−5 : 6242844 : 2) (−6 : 3882000 : 1) (−2 : 533610000 : 7) (−8 : 24126624 : 1) (−8 : 36660000 : 3) (9 : 109597875 : 1) (−7 : 5861808750 : 12) (−14 : 315937776 : 5) (−16 : 665016000 : 1) (−16 : 4284072000 : 11) (17 : 161975250 : 8) (−19 : 1880058375 : 9) (−22 : 1530438000 : 7) (−5 : 57189896049 : 23) (1 : 35243413449 : 23) (8 : 31202028000 : 27) (−28 : 4389294000 : 3) (28 : 9383852016 : 1) (−34 : 5059494000 : 9) (−32 : 470512538496 : 35) (35 : 180289347084 : 26) (36 : 7110310416 : 17) (37 : 23454791250 : 8) (−23 : 596214340500 : 38) (−3 : 342688879500 : 38) (38 : 77478505104 : 21) (48 : 717524088000 : 37) (−49 : 10998894375 : 9) (−49 : 392578147500 : 34) (6 : 581716422000 : 49) (49 : 67438132125 : 11) (49 : 2138095777356 : 54) (−57 : 63805462875 : 7) (−66 : 267313590000 : 31) (28 : 1353422994000 : 67) (16 : 1587115560000 : 69) (−56 : 7785845340000 : 71) (63 : 8836674346500 : 82) (−84 : 81032490000 : 19) (−86 : 625679595024 : 37) (34 : 4067631834000 : 91) (−94 : 832443162000 : 39) (96 : 1871111088000 : 49) (103 : 1847241290466 : 8) (112 : 813489768000 : 53) (118 : 2176373034000 : 27) (122 : 3699882582000 : 13) (138 : 4435761990000 : 67) (142 : 5975166090000 : 3) (48 : 29687227791936 : 151) (−159 : 16854964966125 : 89) (−160 : 15328820639616 : 87) (−166 : 1313381190000 : 31) (14 : 97213816854000 : 171) (−194 : 79827050490000 : 129) (−197 : 1653113143500 : 42) (116 : 189906187746000 : 199) (224 : 37496418960000 : 41) (−160 : 991032738098304 : 241) (261 : 67888502075625 : 49) (82 : 274738571910000 : 263) (−282 : 98369188710000 : 137) (−288 : 21677436528000 : 73) (31 : 916644696453750 : 304) (−373 : 229382096843625 : 13) (304 : 5548257960696000 : 421) (−432 : 10486484580936000 : 427) (434 : 255082089318000 : 111) (−450 : 98875785532656 : 79) (−466 : 201416233958544 : 69) (−512 : 282408676032000 : 77) (512 : 1070937607872000 : 23) (544 : 850319119056000 : 131) (547 : 236968082958750 : 148) (−584 : 121109802660000 : 119) (−588 : 14108863810734000 : 463) (−644 : 1291528975506000 : 219) (−676 : 10806526941210000 : 441) (−779 : 447028109793375 : 169) (−838 : 625449678810000 : 183) (−102 : 118056626949690000 : 907) (−942 : 1453582876746000 : 217) (544 : 135939197462640000 : 1121) (1136 : 132853147300440000 : 749) (1324 : 49891361179590000 : 91) (−1459 : 29607407239325625 : 169) (−1070 : 2753071842509719536 : 1761) (2784 : 336581829779760000 : 731) (3082 : 1481150126753418000 : 433) (3424 : 9125200546325424000 : 2151) (−4912 : 5046733520778840000 : 417) (−6601 : 8278160497833371250 : 1816) (−8378 : 5588322854306010000 : 1673)
# w^2 = P(s0,s1; t0,t1): one term per line, exponents of s0 s1 t0 t1 and the coefficient 0 4 2 2 16 0 4 3 1 8 0 4 4 0 1 1 3 1 3 552 1 3 2 2 138 1 3 3 1 -924 1 3 4 0 114 2 2 0 4 81 2 2 1 3 -9528 2 2 2 2 18078 2 2 3 1 -11766 2 2 4 0 3349 3 1 0 4 720 3 1 1 3 13398 3 1 2 2 -21624 3 1 3 1 1638 3 1 4 0 5700 4 0 0 4 1600 4 0 1 3 1360 4 0 2 2 -3711 4 0 3 1 -1700 4 0 4 0 2500 # split rulings: 0 a b is s = (a:b), 1 a b is t = (a:b) 0 0 1 0 1 0 0 1 1 0 1 4 0 3 -1 0 1 23 0 5 -2 1 1 -4 1 0 1 1 1 1 1 1 0 1 4 -3 1 13 14 1 10 9 # split (1,1)-curves a b c d: a s0 t0 + b s0 t1 + c s1 t0 + d s1 t1 = 0 50 -49 -1 0 53 -52 -1 0 6 -3 3 -4 66 -73 -6 3 10 4 5 -3 8 -10 1 1 1 1 -2 0 25 -38 -1 0 5 -6 1 0 34 -35 2 0 125 -121 -20 16 10 -5 1 -6 5 20 3 -6 26 -19 8 0 2 -5 3 0 45 18 1 -20 78 -111 5 0 84 -96 7 5 6 1 -3 0 94 -107 1 12 10 17 6 -3 48 -51 8 9 26 35 1 0 95 -100 13 6 13 -14 23 0 50 -85 -65 78 46 -32 13 4 6 15 2 -8 152 -172 -19 39 75 -96 1 20 50 -56 3 0 4 3 -148 144 25 28 -23 0 154 -5 7 -2 146 -181 23 12 22 13 17 -52 34 7 -10 0 7 5 84 -96 750 -771 5 36 75 93 -31 -9 20 134 16 -5 80 -25 2 -24 125 -55 2 104 50 59 -37 -10 12 48 11 -36 375 -305 30 24 6 5 -12 32 82 -89 53 -156 312 -320 -24 39 60 114 15 -65 35 16 5 -26 52 100 1 -65 28 -254 35 15 259 70 -7 -12 63 -180 91 26 7 -455 8 0 250 19 -115 0 75 -375 100 32 250 65 11 -2 568 -764 52 39 1422 -36 -9 -13 1400 356 -7 -13 874 -69 -19 -36 # nodes (s0 s1 t0 t1) 0 1 0 1 1 1 1 1
(−1 : 3744 : 1) (0 : 1587 : 1) (1 : 5376 : 1) (1 : 74475 : 2) (−1 : 105600 : 3) (−4 : 724101 : 3) (−1 : 408249 : 4) (4 : 5875125 : 3) (−2 : 527751 : 5) (1 : 2024064 : 5) (3 : 693216 : 5) (−7 : 47808000 : 1) (−1 : 3861504 : 7) (1 : 6276000 : 7) (−9 : 135759264 : 1) (−9 : 13958400 : 7) (−7 : 19855104 : 9) (−1 : 10293024 : 9) (−11 : 310015104 : 1) (−5 : 1135872 : 11) (5 : 72825312 : 11) (11 : 407733024 : 5) (−11 : 124580736 : 17) (1 : 156268800 : 17) (9 : 352095744 : 17) (17 : 2287334400 : 9) (18 : 2550285375 : 1) (−19 : 197232000 : 17) (−21 : 332614656 : 19) (−23 : 3672787200 : 9) (−23 : 2247717504 : 13) (−23 : 1893610125 : 14) (−23 : 319166400 : 19) (−23 : 526148736 : 21) (−13 : 67085184 : 23) (−10 : 134792463 : 23) (−9 : 256909056 : 23) (−1 : 425510400 : 23) (23 : 5586920064 : 19) (−1 : 594548064 : 25) (25 : 10730197632 : 13) (−26 : 607812975 : 23) (13 : 2566302816 : 27) (27 : 12933081600 : 19) (−23 : 2244873600 : 29) (7 : 2616638400 : 29) (7 : 2924874411 : 30) (−31 : 1037822016 : 27) (−23 : 2515989504 : 33) (19 : 3378648000 : 33) (−35 : 15554639232 : 17) (−9 : 2372133504 : 35) (−23 : 2163725184 : 37) (−23 : 1624812000 : 39) (7 : 6962438400 : 39) (−23 : 24747264 : 41) (13 : 12598992000 : 41) (41 : 44435664000 : 37) (−43 : 16712486400 : 29) (−9 : 8010508800 : 47) (−3 : 8768419200 : 49) (−19 : 8125489575 : 52) (−53 : 41700681216 : 35) (55 : 233647772928 : 7) (−61 : 15199874400 : 53) (−37 : 13115037600 : 61) (−35 : 6021320448 : 61) (−27 : 5126016000 : 61) (−47 : 41164870656 : 73) (−41 : 2660066400 : 73) (38 : 124169053239 : 73) (44 : 17299880325 : 73) (29 : 179550034125 : 78) (−94 : 1209382260801 : 29) (−68 : 191871445125 : 99) (−101 : 125276659200 : 83) (107 : 346617821319 : 108) (−73 : 251641051200 : 109) (−137 : 495682137600 : 111) (−108 : 1126724495925 : 139) (145 : 11622166553088 : 89) (101 : 2188689621504 : 173) (−23 : 1633454222400 : 179) (−163 : 6325350796800 : 229) (−97 : 6258625632000 : 271) (−23 : 13936317283200 : 309) (103 : 45696933942336 : 317) (115 : 54059349222528 : 327) (335 : 350145933099072 : 117) (−341 : 30927276432000 : 263) (−355 : 39463911978048 : 271) (−261 : 41534094134400 : 383) (−119 : 37929866743296 : 401) (−69 : 44801378294400 : 407) (−187 : 23057380464000 : 441) (−415 : 263993731694253 : 574) (943 : 10568183173840800 : 881) (−437 : 205366189075200 : 971) (−977 : 2088579328416000 : 911) (349 : 10967231987019264 : 1285) (−1261 : 18676485715872000 : 1523) (−341 : 12855022842225600 : 1673) (225 : 84938207232463872 : 2521) (397 : 269091418634497536 : 3253) (−2691 : 450174544417092000 : 3563)
149 points: y^2 = 3620389893120*x^7 - 4613852710656*x^6 + 2505007146096*x^5 - 750705993891*x^4 + 134025906966*x^3 - 14243265036*x^2 + 833395710*x - 20682900 149 points: y^2 = 115500000*x^7 + 45800625*x^6 - 581214750*x^5 + 79392225*x^4 + 664893840*x^3 - 53009664*x^2 - 182730240*x + 27878400 149 points: y^2 = -1493856*x^7 - 40786823*x^6 + 273284956*x^5 + 17685468252*x^4 + 188697104416*x^3 + 738130188400*x^2 + 870263361984*x + 455779768896