Datasets:
The dataset viewer is not available for this dataset.
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Quantum Ground States — Exact Representations to 320 Qubits
Exact quantum ground states of disordered transverse-field Ising chains, from 20 to 320 qubits, each paired with its local quantum representation and a physical label. A benchmark for learning properties of quantum states at a scale no state-vector simulator can reach — every example is exact ground truth.
Benchmark
Predict the physical observable Σᵢ ⟨ZᵢZᵢ₊₁⟩ from a representation of the ground state. Two
representations are compared under one protocol (kernel-ridge regression; error normalised so that
chance = 1.0; lower is better):
- Quantum — the local single-qubit measurements of the ground state (the 1-RDM).
- Classical — the parameters that generate the chain (the disorder fields).
| qubits | quantum representation | classical representation |
|---|---|---|
| 20 | 0.10 | 0.48 |
| 40 | 0.11 | 0.45 |
| 80 | 0.13 | 0.53 |
| 160 | 0.21 | 0.61 |
| 320 | 0.40 | 0.80 |
The quantum representation is 2–5× more accurate at every scale and remains well clear of
chance at 320 qubits. The sample-efficiency advantage holds past the point where the full quantum
state — 2³²⁰ ≈ 10⁹⁶ amplitudes — can no longer be written down. Mean over 3 seeds; standard
deviations in verification.json.
What's in it
| field | type | description |
|---|---|---|
n_qubits |
int | chain length: 20, 40, 80, 160, 320 |
fields |
list[float] | disorder parameters hᵢ ~ U(0.2, 2.0), length n_qubits |
quantum_features |
list[float] | local 1-RDM ⟨Xᵢ⟩, ⟨Yᵢ⟩, ⟨Zᵢ⟩, length 3·n_qubits |
label |
float | Σᵢ ⟨ZᵢZᵢ₊₁⟩, an exact physical observable |
750 examples, 150 per scale.
Exact ground truth at 320 qubits
The transverse-field Ising chain is exactly solvable by the Jordan–Wigner / Bogoliubov–de Gennes
transformation: its ground state is fixed by an n × n matrix, not a 2ⁿ state vector. Every
representation and label here is exact, validated to 10⁻¹⁴ against full diagonalisation. That is
what makes a clean benchmark possible at hundreds of qubits — exact answers to measure against, no
approximation.
The representation
quantum_features are the local 1-RDM — single-qubit expectation values of the ground state. This
is the representation that scales:
- Hardware-native. Single-qubit measurements, the standard readout of a quantum processor — validated on IQM and IBM hardware to a few percent. No full-state tomography.
- Stable under scale. Representations that compare whole states by global overlap lose their signal as the system grows; local representations do not. This is why the benchmark runs to 320 qubits.
Produced by the ReLab quantum data engine — the same representation it generates from a customer's own systems.
Quickstart
from datasets import load_dataset
ds = load_dataset("SiriusQuantum/quantum-ground-states-320-qubits")["train"]
X, y = ds["quantum_features"], ds["label"] # filter by ds["n_qubits"] for the scaling curve
Scope
- A benchmark of representation and sample efficiency against exact ground truth; the exactly-solvable model is what enables verification at this scale.
- Results are reported under a fixed protocol; the headline is the consistent gap between the quantum and classical representations across scales.
- The representation is computed by the exact solver here and is identical to what a quantum processor measures on-device.
Reproducibility
Every example regenerates deterministically from recipe.json (engine commit, seed, generator).
verification.json carries the benchmark result. The solver is validated to 10⁻¹⁴ against exact
diagonalisation for n ≤ 12.
References
- Lieb, Schultz, Mattis 1961 — exact free-fermion solution of the spin-½ XY / TFIM chain.
- Huang, Broughton, Mohseni, et al. 2021 — arXiv:2011.01938 — projected (local) quantum kernels.
- Huang, Kueng, Torlai, Albert, Preskill 2022 — arXiv:2106.12627 (Science) — provably efficient ML for quantum many-body ground-state properties.
- Lewis, Huang, Tran, et al. 2024 — Nat. Commun. 15:895, doi:10.1038/s41467-024-45014-7 — O(log n) sample complexity for ground-state properties.
- Thanasilp, Wang, Cerezo, Holmes 2022 — arXiv:2208.11060 — concentration of global quantum kernels.
Citation
@misc{siriusquantum2026qgs320,
title = {Quantum Ground States: Exact Representations to 320 Qubits},
author = {{Sirius Quantum}},
year = {2026},
publisher = {Hugging Face},
howpublished = {\url{https://huggingface.co/datasets/SiriusQuantum/quantum-ground-states-320-qubits}}
}
Produced with the ReLab quantum data engine, Sirius Quantum — https://github.com/Sirius-Quantum
License
CC-BY-4.0 — use is permitted with attribution to Sirius Quantum.
- Downloads last month
- 52