QUBOLib benchmark corpus

QUBOLib benchmark corpus — v0.2.4-data+8 QUBOLib is a curated corpus of binary quadratic optimization benchmarks with models, reference solutions, submitted solutions, and source provenance. This dataset contains 9294 instances across 7 collections. It preserves the exact qubolib.tar.gz distributed through the GitHub data release, not a software source archive. File identity and schema Compressed size: 729302742 bytes. SHA-256: 2470f3b1b8b3ba4f718987883f56f8a47554a54e8735a262b268881dd1904bfc. Julia artifact tree SHA-1: db23630c10bf992c60033093eb46067fad78ccaa. Archive members: archive.h5 and index.db. Build: Julia 1.10.11, QUBOLib commit 37d1d0a2a524750fadbb5d3de484946a548deda8. Data release target: QUBOLib commit 37d1d0a2a524750fadbb5d3de484946a548deda8. Inventory audited: 2026-10-31. index.db is the SQLite catalog of collections, instances, submissions, and solution records. archive.h5 holds models, solution payloads, and source provenance compressed with chunked HDF5 deflate (level 3). Solution records distinguish source objectives, evaluated QUBO values, bounds, validation status, and source feasibility. Source LP content is stored only when its blob is at most 1,000,000 bytes; larger sources retain their URL, size, and SHA-256 metadata. The schema is defined by QUBOLib at the release commit. The machine-readable provenance inventory uses metadata schema version 1. Collections, attribution, and modifications Each of the following collections has verified redistribution terms (CC BY 4.0 or MIT). Retain the attribution and modification notice for every collection used. Upstream creators retain credit for their benchmark instances; QUBOLib curates and converts them for this corpus. arXiv-1903-10928-3r3x 3200 instances. Equation Planting: A Tool for Benchmarking Ising Machines; source. License: CC-BY-4.0; rights evidence. Mirrored source archive; SHA-256 a1b13f38112276f010e5559868eeeb3816b43b5c6753d1f44d49acb910ad13b0. QUBOLib parses the Qubist source files and stores the converted models in the artifact's HDF5/SQLite representation; source-paper attribution is retained. arXiv-1903-10928-5r5x 307 instances. Equation Planting: A Tool for Benchmarking Ising Machines; source. License: CC-BY-4.0; rights evidence. Mirrored source archive; SHA-256 da773fab84c4e6e621efdf090a2fda9fa43420be4c2f2010a6d8fe88b1f34045. QUBOLib parses the Qubist source files and stores the converted models in the artifact's HDF5/SQLite representation; source-paper attribution is retained. arXiv-2103-08464-3r3x 2300 instances. 3-Regular 3-XORSAT Planted Solutions Benchmark of Classical and Quantum Heuristic Optimizers; source. License: CC-BY-4.0; rights evidence. Mirrored source archive; SHA-256 a99b22c02aac03446589b7d366354bc448e6a8bae3e67120ac4bcc821601735b. QUBOLib parses the Qubist source files and stores the converted models in the artifact's HDF5/SQLite representation; source-paper attribution is retained. qplib 23 instances. QPLIB: a library of quadratic programming instances; source. License: CC-BY-4.0; rights evidence. Source commit: 4b681a263d9c2cb21e83f08e1b3232361dcedd5a. Mirrored source archive; SHA-256 896b2e1c5e5e78e7b89674cb6bc84a0f6e05a0e36adddeaaa744472733517672. QUBOLib redistributes a processed subset: it selects the 23 QBB instances, parses each QPLIB model with the specification's 0.5 quadratic objective scaling, uses LP variable names to map SOL entries, cross-references upstream MINLPLib, BiqMac, and MQLib optimality bounds (12 proven optimal, 11 open), and stores the resulting models and solutions in HDF5 and SQLite. qoblib 433 instances. The Quantum Optimization Benchmarking Library; source. License: CC-BY-4.0; rights evidence. Source commit: 16a166ee67c24c112551c803aab5394743b815b5. QUBOLib parses authoritative QS/QUBO models and stores models, provenance, and solution records in HDF5/SQLite. Portfolio positions are converted to canonical ordered copy slots and reconstructed capital/budget slack, not original solver bitstrings. minlplib 29 instances. MINLPLib—a collection of test models for mixed-integer nonlinear programming; source. License: CC-BY-4.0; rights evidence. Source version: 9472b011. Mirrored source archive; SHA-256 0154d83f82b2bfb559b8b3542bad2379f5e3b68c4390997085c8f7ce2d3e33c4. QUBOLib selects the 29 pure binary unconstrained QUBO instances in MINLPLib not already included in QPLIB (2 autocorr_bern, 20 chimera, and 7 sporttournament06..20 instances), parses POLIP (.pip) models and 42 .p*.sol solution records, factors exact 0.1 scaling where applicable, and stores the resulting models and solutions in HDF5 and SQLite. mqlib 3002 instances. What Works Best When? A Systematic Evaluation of Heuristics for Max-Cut and QUBO; source. License: MIT; rights evidence. Source commit: 585496274af5abb0849d0d47e135496b4688680b. Mirrored source archive; SHA-256 af63f5fe3951b6e287bdb6a83fbcf1765f087c208ce832d2229d966ddeeb6c90. QUBOLib imports 3002 benchmark instances from MQLib across all 18 source families (all 231 standard literature benchmark instances from MQLib data/standard.csv, including all 21 Palubeckis p3000..p7000 instances, plus 2771 Tier 2 instances across SteinLib, Rudy, Color02, BiqMac, ImageSegmentation, NetworkX, TSP, CrossGrid, Culberson, MarketSplit, MaxCut, Pardalos, Fortran, SpinGlass, GlasgowIsomorphism, and Dharwadker), converts Max-Cut and master-node QUBO edge lists to canonical :bool/:max QUBO models, attaches 231 validated heuristic bitstring solutions and 3022 value-only heuristic benchmark records across 37 heuristic solver submissions from MQLib data/results.zip, and records the 394 remaining ultra-dense Tier 2 instances in tier2_deferred_inventory. QOBLIB, QPLIB, MINLPLib, and MQLib solution inventory and preprocessing enrichments In v0.2.4-data+8, all 9294 instances across all 7 collections include certified QUBOPreprocessing.jl v0.1.5 reduction metadata (n_strong, m_strong, n_weak, m_weak) and coercive local-field diagnostics (coercive_fields), along with 343 instances solved completely to zero residual variables (333 on MQLib in weak mode plus 6 in strong mode, 6 on QOBLIB, 3 on MINLPLib, and 2 on QPLIB including QPLIB_5721) and 51 certified repaired solution records (32 on QOBLIB, 16 on MQLib across 14 instances, and 3 on MINLPLib). Reproduction and use For byte-exact reproduction, download qubolib.tar.gz from the related GitHub data release or this dataset record, run sha256sum qubolib.tar.gz, and compare the value above. Extract to a new empty directory. Before opening the files with a database application, compute the Julia tree hash using import Pkg; println(bytes2hex(Pkg.GitTools.tree_hash(ARGS[1]))) with that directory as the argument. Citation Cite the version DOI for the exact corpus used and cite each source collection used. Cite the concept DOI when referring to the corpus across versions. For the JuliaQUBO methods and software ecosystem, also cite QUBO.jl: a Julia ecosystem for quadratic unconstrained binary optimization.