HamLib

HamLib is a problem-instance library, not a benchmark protocol: it defines no circuits and produces no score. It earns a place in an application-level catalog because it has become a standard source of inputs for benchmarks that do: most visibly the QED-C Application-Oriented Benchmarks, whose Hamiltonian-simulation benchmark draws its Hamiltonians from HamLib. Assembled by Sawaya and 15 co-authors across Intel Labs, Lawrence Berkeley National Laboratory, Sandia, NASA Ames, Oxford, and other institutions, the dataset packages more than 1.5 million pre-encoded qubit Hamiltonians for benchmarking quantum algorithms and hardware (arXiv:2306.13126; published in Quantum 8, 1559 (2024)).

How it works

The library spans condensed-matter models (transverse-field Ising, Heisenberg, Fermi–Hubbard, Bose–Hubbard), electronic- and vibrational-structure chemistry, and combinatorial optimization (MaxCut, Max-k-SAT, Max-k-Cut, QMaxCut, traveling salesperson), at sizes from 2 to 1000 qubits. Every instance is already mapped to qubit operators (with Jordan–Wigner, parity, and Bravyi–Kitaev fermionic mappings and unary/Gray/binary encodings where relevant) and distributed as zipped HDF5 files from a NERSC-hosted portal. A benchmark built on HamLib picks instances, runs an algorithm or device on them, and applies its own figure of merit: the QED-C Hamiltonian-simulation benchmark, for example, uses five HamLib models (TFIM, Heisenberg, Fermi–Hubbard, Bose–Hubbard, Max3SAT).

Strengths and limitations

HamLib’s value is standardization: it removes instance-preparation work and makes results reproducible and comparable across papers, the role a shared instance library also plays for the QB GSEE Benchmark in chemistry. The flip side: “the HamLib benchmark” is a common misnomer, since the scoring protocols live elsewhere. Dataset versioning is informal (the paper notes “small changes in v1.1 of dataset”), so instance-level reproducibility across revisions is worth checking, and generic web searches collide with the unrelated Hamlib amateur-radio library.

Notable results

The dataset remains live at NERSC as of mid-2026, and QED-C work keeps building on it: a cross-model Hamiltonian-simulation benchmarking framework (arXiv:2409.06919) and a 2025 observable-estimation framework (arXiv:2504.09813) both draw their problem instances from HamLib.