sum4cubes / README.md
kzoacn's picture
Upload generation and checking scripts
5358a7c verified
|
Raw
History Blame Contribute Delete
1.9 kB
metadata
license: cc-by-4.0
task_categories:
  - tabular-regression
language:
  - en
pretty_name: Sum of Four Cubes Certificates

Sum of Four Cubes Certificates

This dataset contains explicit certificates for

n = x1^3 + x2^3 + x3^3 + x4^3

for every integer 0 <= n <= 1,000,000,000.

Each row has the form

123 = (x1)^3 + (x2)^3 + (x3)^3 + (x4)^3

The files are sharded by intervals of length 10,000,000, with the endpoint 1,000,000,000 appended to the last shard:

cubes-00.txt   0 <= n <= 9,999,999
cubes-01.txt   10,000,000 <= n <= 19,999,999
...
cubes-99.txt   990,000,000 <= n <= 1,000,000,000

The generation code uses a congruence-filtered generalized Pell search. For the first successful search parameter d, it records the representation with minimal height

H = max(|x1|, |x2|, |x3|, |x4|)

among the parity-compatible Pell representatives returned by the reference solver.

Reproduce

Generate shards with 15 worker processes:

sage -python scripts/generate_hf_dataset.py \
  --start 0 --end 1000000000 \
  --workers 15 \
  --chunk-size 10000 \
  --max-in-flight 30 \
  --out-dir generated/sum4cubes \
  --overwrite \
  --merge-final-singleton \
  --progress-every 10000000

For long runs, restart the generator from shard boundaries with --resume. For example:

sage -python scripts/generate_hf_dataset.py \
  --start 80000000 --end 129999999 \
  --workers 15 \
  --chunk-size 10000 \
  --max-in-flight 30 \
  --out-dir generated/sum4cubes \
  --resume

Run a lightweight post-generation sanity check:

python3 scripts/check_generated_shards.py \
  --dir generated/sum4cubes \
  --start 0 --end 1000000000

The generator verifies each identity using exact integer arithmetic before writing a row. The sanity checker verifies shard coverage and exact four-cube identities for boundary rows.