n1-landscape

4d 𝒩=1 landscape: regenerated records

Every record of the enumeration of four-dimensional 𝒩=1 theories obtained from seed gauge theories with a simple gauge group by adding superpotential terms one at a time, with the superconformal R-symmetry fixed by a-maximization, decoupled operators removed by flip fields, and each candidate tested with its superconformal index (arXiv:2408.02953, arXiv:2510.19136). The records were computed again from the seeds with the package landscape (package 0.6.0 at commit 8046126 of that repository, native extension 0.12.0; pipeline_commit records the development commit a4b6e37 of the same code), level by level until a level has no input. This version supersedes the dataset's first one (commit ba87844, computed with the package 0.5.2): every seed is computed again with the version above, and the earlier records remain in the dataset's history.

In progress (2026-10-12). 85 of the 150 seeds are in the dataset: 82 completed (the enumeration reached a level with no input), 3 cut before a level of more than 1,000,000 inputs, and 0 deferred (stopped at an open level, by memory or by a limit of the package; to be resumed with a later version of the package). 65 seeds are still being computed; they are added when ready. The dataset holds 6,429,384 records, of which 313,727 are fixed points.

The published database at qft.kaist.ac.kr/landscape, which arXiv:2408.02953 links to, is a different and unchanged collection; its ids are not the ids used here.

Files

• records/<seed>.parquet: one row per record of the seed, in the order of the enumeration (level, then input order).
• fixed_points.parquet: every fixed point of the seeds in the dataset with the short columns (seed, ordinal, level, a, c, a_float, c_float, ratio, rational, flavor_rank and the display strings), sorted by a_float; a table derived from the record files, for searches by central charge.
• seeds.parquet: one row per seed of the campaign (in the campaign order: gauge rank, then name), with its status, the level reached, its counts per level, where and why it stopped (stopped_level, stop, next_level_inputs_counted; for a cut seed the count of distinct inputs stops once past 1,000,000), and its range of ids.

Ids

A record's id is v2-<n> with n = id_offset + ordinal + 1, id_offset from seeds.parquet and ordinal the record's position in its seed file (from 0). The ids follow the campaign order of the seeds in the dataset and are provisional: until the paper that uses this dataset they shift when a seed earlier in the order is added or resumed.

-- DuckDB
SELECT 'v2-' || (s.id_offset + r.ordinal + 1) AS id, r.*
FROM 'hf://datasets/cms1308/n1-landscape/records/*.parquet' r
JOIN 'hf://datasets/cms1308/n1-landscape/seeds.parquet' s USING (seed)
WHERE r.fixed_point;

Columns of records

• ordinal, seed, level; verdict; fixed_point (true when verdict is consistent and the record is not a duplicate); duplicate_of (the input key of the earlier record with the same fixed point) and duplicate_of_ordinal; parent_ordinal (the record of the level before that this one deforms).
• a, c: the central charges as stored, an exact rational p/q or a 30-digit decimal (rational says which); a_float, c_float, ratio (= a/c) for sorting; flavor_rank.
• The record itself as JSON strings (record schema 36.0 of the package): theory (gauge nodes [type, rank], fields as Dynkin labels, superpotential terms as exponent vectors, field names), charges, flavor, index (terms [1000E, 2j_2, flavor exponents, coefficient] through t^t_order), operators, canonical, provenance, analysis.
• canonical_hash, identity, fixed_point_key, input_key: hashes used by the enumeration to identify theories and fixed points.
• Display strings (LaTeX): gauge, fields, superpotential_tex, rcharges, index_t6_tex (the unrefined reduced index to t^6), relevant_tex, marginal_tex, flipped_tex, decoupled_tex, deformation.
• dim3: the t^6 coefficient of the unrefined scalar part of the reduced index (from analysis).
• pipeline_commit, computed_at, schema_version, source.

The verdicts are those of the package (landscape/amax.py, landscape/post.py): consistent, inconsistent-index, non-positive-r-charge, no-r-symmetry, r-charge-undetermined, no-local-maximum, index-not-computed, post-processing-error, hofman-maldacena-violation, negative-central-charge. An index-not-computed record gives its cause in provenance.not_computed: "expansion-order" (the expansion order above 100, or an R-charge too close to 0 or 2) or "work-bound" (the expansion formed more than 10^10 product terms; "bound" records it).

analysis.extra_supercurrents (Σ_q max(c_q, 0), c_q the coefficient of t^7(y + 1/y) at flavor charge q in the reduced index) and analysis.nonmanifest_symmetry (from the t^6 coefficient of the flavor-refined reduced index) are signals in the index computed by the package. They are not established supersymmetry enhancement or symmetry; the 𝒩=2 fixed points of the published database were established by separate checks of the index and its 𝒩=2 limits.

License and citation

CC BY 4.0. If you use these records, please cite arXiv:2408.02953 and arXiv:2510.19136, and this dataset.