Elliptic curves by rank
Elliptic curves over Q with explicit Mordell–Weil generators, in three separate collections.
Every curve is given by its global minimal model [a1,a2,a3,a4,a6], and every generator is a point
[x,y] on that model. Coordinates and conductors are exact; read them as integers and rationals, not as
floating point numbers. The three collections are kept apart, and a curve can appear in more than one of them (for
example, the small rank 4 curves of the census are also among the search curves). To read a curve in PARI/GP:
E = ellinit([a1,a2,a3,a4,a6]);
1. All curves of rank ≥ 3 with |Δ| ≤ 1013
These files list every elliptic curve over Q of rank at least 3 whose minimal model has discriminant Δ and invariant c4 with |Δ| ≤ 1013 and c4 ≤ 1.44·1012: 2,302,875,461 of the 78,220,212,957 curves in that region. The bound on c4 is needed because |Δ| ≤ 1013 alone allows arbitrarily large c4 near c43 = c62; of the 65,652 curves in Cremona's tables with |Δ| ≤ 499,999, only three have larger c4. There is one file per rank, sorted by conductor; since the conductor divides the discriminant, every conductor here is at most 1013. The region is not complete by conductor: curves of small conductor can have a large discriminant.
| file | rank | curves | E | S | L | A | smallest conductor |
|---|---|---|---|---|---|---|---|
census_rank3.tsv.zst (60.7 GB) | 3 | 2,112,313,522 | 2,111,695,943 | 616,732 | 572 | 275 | 5077 [0,0,1,-7,6] |
census_rank4.tsv.zst (5.2 GB) | 4 | 183,192,860 | 183,192,833 | 27 | 234446 [1,-1,0,-79,289] | ||
census_rank5.tsv.zst (214.9 MB) | 5 | 7,276,292 | 7,276,292 | 19047851 [0,0,1,-79,342] | |||
census_rank6.tsv.zst (3.1 MB) | 6 | 92,666 | 92,666 | 5187563742 [1,1,0,-2582,48720] | |||
census_rank7.tsv.zst (5.7 kB) | 7 | 121 | 121 | 536670340706 [1,0,1,-14733,694232] | |||
| all | 2,302,875,461 | 2,302,257,855 | 616,759 | 572 | 275 |
What the ranks rest on. rank_lo is the number of independent points listed, an unconditional lower
bound, and rank_hi is the 2-descent upper bound, also unconditional. The status column says how the
listed rank is established:
| status | bounds | the rank | curves |
|---|---|---|---|
| E | rank_lo = rank_hi | proven: the listed points and the 2-descent bound agree | 2,302,257,855 |
| S | rank_lo = rank_hi − 1 | equals rank_hi if the Tate–Shafarevich group is finite (its 2-primary part is enough); unconditionally rank_lo or rank_hi | 616,759 |
| L | [3,5] | 3 if the Birch and Swinnerton-Dyer conjecture holds (the analytic rank is 3, assuming GRH); unconditionally 3, 4 or 5 | 572 |
| A | [0,3] or [1,3] | 3 if the Birch and Swinnerton-Dyer conjecture holds (the analytic rank is 3); fewer than three points are known | 275 |
So the rank is proven exactly for 99.973% of the listed curves. The S curves at rank 3 also have analytic rank exactly 3, assuming GRH.
Completeness. Every curve in the region that is not listed has analytic rank at most 2, assuming the Generalized Riemann Hypothesis for its L-function. Its rank is then at most 2: by the theorems of Gross–Zagier and Kolyvagin when the analytic rank is at most 1, and by the Birch and Swinnerton-Dyer conjecture when it is 2.
Check against Cremona's tables. All 8,561 curves of rank ≥ 3 in John Cremona's database (conductor ≤ 499,999) that lie in the region are listed here with the same rank, and every listed curve of conductor ≤ 499,999 is one of them.
Format
Tab-separated with a header line, compressed with zstd:
zstdcat census_rank3.tsv.zst | head. The rank 3 file holds 2.1 billion lines; read it as a stream
(for example pandas.read_csv(..., sep="\t", compression="zstd", chunksize=10**7)) rather than all at once.
| column | meaning |
|---|---|
conductor | the conductor N, exact integer |
a_invs | [a1,a2,a3,a4,a6] of the minimal model |
rank | the rank: proven for status E, the value in the status table otherwise |
rank_lo | number of independent points listed (unconditional lower bound) |
rank_hi | 2-descent upper bound (unconditional) |
status | E, S, L or A, as above |
torsion_order | order of the torsion subgroup |
generators | rank_lo independent points [x,y] on the listed model |
Torsion orders of the listed curves: trivial 2,302,410,005; order 2: 464,750; order 3: 646; order 4: 60.
2. Curves of rank 4 to 15 from searches
Every curve here has proven rank. For each one the listed generators are verified to lie on the curve and to be independent, which gives the rank as a lower bound, and a 2-descent upper bound matches it. Every model is the minimal model, and no curve appears twice.
| file | rank | curves | smallest log N | smallest naive height | smallest Faltings height | smallest log abs(disc) |
|---|---|---|---|---|---|---|
rank_4.tsv.gz (24.2 MB) | 4 | 581,559 | 12.364981 | 21.574288 | -0.173133 | 13.058128 |
rank_5.tsv.gz (21.7 MB) | 5 | 421,550 | 16.762465 | 24.317973 | 0.083690 | 16.762465 |
rank_6.tsv.gz (90.0 MB) | 6 | 1,396,828 | 22.369530 | 30.376041 | 0.582833 | 22.643449 |
rank_7.tsv.gz (253.9 MB) | 7 | 3,681,098 | 26.670318 | 35.779031 | 1.036540 | 28.235073 |
rank_8.tsv.gz (237.1 MB) | 8 | 2,883,871 | 33.151079 | 41.826383 | 1.511528 | 33.644948 |
rank_9.tsv.gz (331.9 MB) | 9 | 3,301,005 | 38.007861 | 47.863736 | 1.982707 | 39.095558 |
rank_10.tsv.gz (246.2 MB) | 10 | 1,665,924 | 43.767868 | 54.348977 | 2.510505 | 45.376023 |
rank_11.tsv.gz (408.1 MB) | 11 | 2,268,936 | 51.246420 | 61.185893 | 3.041194 | 51.246420 |
rank_12.tsv.gz (135.6 MB) | 12 | 683,638 | 57.764522 | 68.672711 | 3.722319 | 59.188359 |
rank_13.tsv.gz (11.9 MB) | 13 | 59,552 | 64.738469 | 75.137429 | 4.245237 | 65.837081 |
rank_14.tsv.gz (680.9 kB) | 14 | 3,315 | 72.303161 | 82.371945 | 4.863211 | 73.720054 |
rank_15.tsv.gz (5.8 kB) | 15 | 26 | 81.565498 | 92.182205 | 5.637976 | 82.746402 |
16,947,302 curves in total. Every figure in the table links to the ten smallest curves of that rank in that category, with their a-invariants, exact conductors and full generators. The four minima in each row are taken independently, so they are generally attained by four different curves. Heights are natural logarithms. The naive height is log max(|c4|3, c62), and abs(disc) is the absolute value of the minimal discriminant. The Faltings height is the unstable one, normalized as hFal(E) = −(1/2) log |Im(conj(ω1) ω2)|, with ω1, ω2 a basis of the period lattice of the minimal model, the normalization used by the Elliptic Curve Rank Leaderboard. It is negative for some rank 4 curves.
Format
Tab-separated with a header line, gzipped, one curve per line, sorted by ascending conductor; zcat, pandas,
R and awk read them directly.
| column | meaning |
|---|---|
rank | the rank, proven |
conductor | the conductor N, exact integer |
log_N | log N |
naive_h | naive height, log max(|c4|3, c62) |
a_invs | [a1,a2,a3,a4,a6] of the minimal model |
generators | independent points generating a subgroup of finite index |
Points may have rational coordinates: 32.5% of these curves have a generator with a denominator, and denominators reach 35 digits. Most conductors exceed 253. No curve appears twice, but some are isogenous to each other (isogenous curves share rank, conductor and L-function): there are 32,651 isogenous pairs, almost all at ranks 6 and 7.
By torsion subgroup
The curves above with a nontrivial torsion subgroup, one file per group, with the same columns plus
torsion (the group) and torsion_generators (points generating the torsion subgroup). Every curve in
these files also appears in its rank_r file.
| file | torsion | curves | ranks | smallest log N by rank |
|---|---|---|---|---|
torsion_Z2.tsv.gz (60.1 MB) | Z/2 | 644,993 | 4 to 9 | 4: 15.298, 5: 19.263, 6: 24.529, 7: 30.739, 8: 36.438, 9: 44.625 |
torsion_Z3.tsv.gz (887.5 kB) | Z/3 | 7,177 | 4 to 9 | 4: 16.239, 5: 21.625, 6: 26.966, 7: 36.085, 8: 42.561, 9: 61.899 |
torsion_Z4.tsv.gz (485.4 kB) | Z/4 | 4,314 | 4 to 6 | 4: 17.737, 5: 22.184, 6: 29.901 |
torsion_Z5.tsv.gz (315.1 kB) | Z/5 | 2,156 | 4 to 6 | 4: 21.613, 5: 29.021, 6: 36.471 |
torsion_Z6.tsv.gz (8.7 kB) | Z/6 | 90 | 4 to 5 | 4: 19.975, 5: 29.321 |
torsion_Z7.tsv.gz (1.6 kB) | Z/7 | 8 | 4 to 4 | 4: 29.804 |
torsion_Z8.tsv.gz (3.9 kB) | Z/8 | 21 | 4 to 4 | 4: 29.638 |
torsion_Z2xZ2.tsv.gz (21.1 kB) | Z/2xZ/2 | 210 | 5 to 6 | 5: 22.340, 6: 29.049 |
torsion_Z4xZ2.tsv.gz (337 B) | Z/4xZ/2 | 2 | 4 to 4 | 4: 22.553 |
torsion_Z6xZ2.tsv.gz (5.6 kB) | Z/6xZ/2 | 25 | 4 to 4 | 4: 31.586 |
Smallest log N by rank and torsion group:
| rank | Z/2 | Z/3 | Z/4 | Z/5 | Z/6 | Z/7 | Z/8 | Z/2xZ/2 | Z/4xZ/2 | Z/6xZ/2 |
|---|---|---|---|---|---|---|---|---|---|---|
| 4 | 15.298 | 16.239 | 17.737 | 21.613 | 19.975 | 29.804 | 29.638 | 22.553 | 31.586 | |
| 5 | 19.263 | 21.625 | 22.184 | 29.021 | 29.321 | 22.340 | ||||
| 6 | 24.529 | 26.966 | 29.901 | 36.471 | 29.049 | |||||
| 7 | 30.739 | 36.085 | ||||||||
| 8 | 36.438 | 42.561 | ||||||||
| 9 | 44.625 | 61.899 |
3. Curves from parametrized families
Curves obtained by specializing parametrized families of elliptic curves (elliptic surfaces over Q(t)). A curve
is included only when its rank is larger than the generic rank of the family it comes from. There is one file per
torsion subgroup, with the same columns as the torsion files above; for trivial torsion torsion_generators is
[]. Every curve has proven rank, is given by its global minimal model, and appears once.
| file | torsion | curves | ranks | smallest log N by rank |
|---|---|---|---|---|
family_trivial.tsv.gz (84.4 kB) | trivial | 134 | 13 to 21 | 13: 77.797, 14: 86.302, 15: 96.221, 16: 107.477, 17: 119.205, 18: 121.679, 19: 137.182, 20: 145.331, 21: 166.030 |
family_Z2.tsv.gz (812.8 MB) | Z/2 | 1,755,871 | 10 to 17 | 10: 56.173, 11: 65.938, 12: 74.055, 13: 85.035, 14: 95.519, 15: 101.026, 16: 131.966, 17: 146.353 |
family_Z4.tsv.gz (90.5 MB) | Z/4 | 350,184 | 5 to 10 | 5: 24.649, 6: 32.690, 7: 40.747, 8: 48.168, 9: 61.548, 10: 78.506 |
family_Z5.tsv.gz (1.9 kB) | Z/5 | 18 | 4 to 4 | 4: 22.139 |
family_Z6.tsv.gz (1.7 kB) | Z/6 | 15 | 4 to 5 | 4: 22.973, 5: 30.341 |
Also here
generators_scatterplots.pdf (1.1 MB): notes on what the Mordell–Weil generators of these curves look like when suitably normalized.