Published November 12, 2026 | Version v3

A Molecular Hamiltonian and Quantum-Hardware Samples Dataset for Sample-based Quantum Diagonalization

Description

A Molecular Hamiltonian and Quantum-Hardware Samples Dataset for Sample-based Quantum Diagonalization

A dataset of 760 molecules across 16 chemical classes, with electronic-structure Hamiltonians (FCIDUMP) and quantum-hardware measurement data (bitstring samples) from two circuit families: LUCJ and SqDRIFT. The samples were measured on IBM Quantum hardware (156-qubit Heron r2 and r3 processors), and the dataset is intended to support training and evaluation of AI models, solvers, and algorithmic workflows for Hamiltonian simulation (e.g. sample-based quantum diagonalization, SQD).

License

Community Data License Agreement – Permissive, Version 2.0 (CDLA-Permissive-2.0). See LICENSE.txt.

Contents

README.md
LICENSE.txt
molecules_info.csv
fcidump.tar.bz2.part000 ... part003
lucj.tar.bz2.part000 ... part024
sqdrift.tar.bz2.part000 ... part024
sqdrift_permutations.tar.bz2
inventory.txt
additional_sqdrift_configs_for_subset_sqdrift.tar.bz2.part000 ... part012
additional_sqdrift_configs_for_subset_sqdrift_permutations.tar.bz2
additional_sqdrift_configs_for_subset_inventory.txt

The four largest archives are split into 500MB parts to work around Zenodo's upload size limits; see "Reassembling the split archives" below.

Reassembling the split archives

Each .partNNN file is a 500MB chunk (zero-padded, 3 digits). After downloading all parts of an archive, concatenate them in order to recover the original .tar.bz2:

cat fcidump.tar.bz2.part* > fcidump.tar.bz2
cat lucj.tar.bz2.part* > lucj.tar.bz2
cat sqdrift.tar.bz2.part* > sqdrift.tar.bz2
cat additional_sqdrift_configs_for_subset_sqdrift.tar.bz2.part* > additional_sqdrift_configs_for_subset_sqdrift.tar.bz2

Shell globbing expands part* in the correct numerically-sorted order, so this works as-is in bash, zsh, or sh. The reassembled file can then be extracted as usual, e.g. tar -xjf fcidump.tar.bz2. sqdrift_permutations.tar.bz2 and additional_sqdrift_configs_for_subset_sqdrift_permutations.tar.bz2 are small enough to upload whole and need no reassembly.

molecules_info.csv

One row per molecule (760 rows), with columns:

Column Description
ID 3-digit molecule ID (001-760), matching the ID used in all file names and archive paths below
Name Molecule name
Formula Molecular formula
SMILES SMILES string
Molecule class One of 16 chemical classes (e.g. Alkanes and cycloalkanes, Esters and lactones, Transition metal complexes, ...)
Active space Active space used for the electronic structure calculation, as (n_electrons e, n_orbitals o)

For the 10 transition-metal complexes (IDs 751-760) the name is a descriptive label (e.g. "Transition metal complex 1").

Core data (all 760 molecules, measured on ibm_aachen, a Heron r3 processor)

  • fcidump.tar.bz2 (split into .part000-.part003) — one FCIDUMP file per molecule (fcidump/<ID>.fcidump), the electronic structure Hamiltonian in the active space given in molecules_info.csv.
  • lucj.tar.bz2 (split into .part000-.part024) — quantum-hardware measurement samples from LUCJ circuits, one file per molecule (shots/lucj/<ID>_shots.npz). See the LUCJ parameters below.
  • sqdrift.tar.bz2 (split into .part000-.part024) — quantum-hardware measurement samples from SqDRIFT circuits, 1,500 files per molecule (shots/sqdrift/<ID>/*.npz). See the SqDRIFT parameters below.
  • sqdrift_permutations.tar.bz2 — the qubit permutation used by each SqDRIFT circuit (shots/sqdrift_permutations/<ID>/<circuit>_metadata.txt, one line per file), required to correctly map each SqDRIFT sample file's qubit columns back to orbital order (see "Loading the samples" below).
  • inventory.txt — full file listing of the four archives above.

LUCJ circuit parameters

Parameter Value
n_reps 1
Number of samples 1.5M

SqDRIFT circuit parameters

Parameter Value
Num. groups 15
num_rand (randomizations per time value) 500
m, time points (Δt = 1) [1.0, 2.0, 3.0]
Num. samples (per circuit) 1,000
Total samples (per molecule) 1.5M

Additional SqDRIFT configurations, 50-molecule subset (measured on ibm_fez, a Heron r2 processor)

  • additional_sqdrift_configs_for_subset_sqdrift.tar.bz2 (split into .part000-.part012)
  • additional_sqdrift_configs_for_subset_sqdrift_permutations.tar.bz2
  • additional_sqdrift_configs_for_subset_inventory.txt

Additional SqDRIFT runs exploring alternative circuit configurations (number of groups, number of randomizations, number of time points) for a subset of 50 molecule IDs. These use the same file layout as the core SqDRIFT archives, with an extra <config> path component identifying which configuration a file belongs to:

shots/sqdrift/<ID>/<config>/*.npz
shots/sqdrift_permutations/<ID>/<config>/*_metadata.txt

Coverage of the 50 molecule IDs is not perfectly uniform across the 12 configurations; see additional_sqdrift_configs_for_subset_inventory.txt for the exact file listing.

SqDRIFT configuration grid

Config Num. groups num_rand m (Δt = 1) Num. samples Total samples
A1 15 1000 [1.0, 2.0, 3.0] 500 1.5M
A2 15 1000 [1.0, 2.0, 3.0, 4.0] 375 1.5M
A3 20 1000 [1.0, 2.0, 3.0] 500 1.5M
A4 20 1000 [1.0, 2.0, 3.0, 4.0] 375 1.5M
B1 15 100 [1.0, 2.0, 3.0] 5000 1.5M
B2 15 100 [1.0, 2.0, 3.0, 4.0] 3750 1.5M
B3 20 100 [1.0, 2.0, 3.0] 5000 1.5M
B4 20 100 [1.0, 2.0, 3.0, 4.0] 3750 1.5M
C1 15 500 [1.0, 2.0, 3.0] 1000 1.5M
C2 15 500 [1.0, 2.0, 3.0, 4.0] 750 1.5M
C3 20 500 [1.0, 2.0, 3.0] 1000 1.5M
C4 20 500 [1.0, 2.0, 3.0, 4.0] 750 1.5M

Every configuration targets 1.5M total samples per molecule.

Loading the samples

Every sample file is a .npz with two arrays:

  • bitstrings: 1D array of fixed-width strings, one character ('0'/'1') per qubit, in little-endian qubit order (as returned directly by the hardware).
  • count: 1D array of integers, the number of times each row of bitstrings was measured. count.sum() is the total sample count for that file.

Only numpy and the Python standard library are needed to load them.

LUCJ

LUCJ samples need no further transformation: the qubit order already matches orbital order, so the bitstrings can be loaded and turned into a boolean configuration matrix directly.

import numpy as np

with np.load("shots/lucj/001_shots.npz") as data:
    bitstrings = data["bitstrings"]   # shape (num_unique,), dtype '<U{num_qubits}'
    counts = data["count"]            # shape (num_unique,), dtype int64

num_qubits = len(bitstrings[0])
num_samples = int(counts.sum())

# Boolean configuration matrix: rows are unique measured configurations,
# columns are qubits in little-endian order; shape (num_unique, num_qubits).
config_matrix = np.array([[c == "1" for c in bs] for bs in bitstrings], dtype=bool)
probabilities = counts / num_samples

SqDRIFT

Each SqDRIFT circuit may measure qubits in a randomized order, recorded as a permutation list in the matching *_metadata.txt file under sqdrift_permutations/. The bitstring columns must be reordered through this permutation before they can be interpreted in a fixed orbital order (e.g. to be merged across different randomizations/time points, or compared with the FCIDUMP active space).

Not every circuit has a permutation file: qubit randomization is only applied when the mode-rebalancing step used to optimize circuit depth succeeds, so some circuits keep the identity ordering and have no corresponding *_metadata.txt file. Treat a missing permutation file as the identity permutation (list(range(num_qubits))) — this is the same fallback used internally to post-process these samples.

import ast
import glob
import os
import re

import numpy as np


def load_sqdrift_samples(sample_npz_path, permutation_txt_path=None):
    """Load one SqDRIFT sample file and permute its columns to orbital order.

    `permutation_txt_path` may point to a file that doesn't exist (or be
    None): a circuit with no mode-rebalancing has no permutation file, which
    is equivalent to the identity permutation.
    """
    with np.load(sample_npz_path) as data:
        bitstrings = data["bitstrings"]
        counts = data["count"].astype(np.int64)

    num_qubits = len(bitstrings[0])

    if permutation_txt_path is not None and os.path.exists(permutation_txt_path):
        with open(permutation_txt_path) as f:
            permutation = ast.literal_eval(f.read().strip())
    else:
        permutation = list(range(num_qubits))

    # The stored bitstrings are little-endian; convert the (big-endian)
    # permutation to the equivalent little-endian column order.
    permutation = np.asarray(permutation)
    permutation_little_endian = (num_qubits - 1) - permutation[::-1]

    # View each fixed-width string as an array of per-character codes, permute
    # the columns, then view back as strings.
    n = len(bitstrings)
    codes = bitstrings.view(np.uint32).reshape(n, num_qubits)
    permuted_codes = np.take(codes, permutation_little_endian, axis=1)
    permuted_bitstrings = permuted_codes.view(f"<U{num_qubits}").ravel()

    config_matrix = permuted_codes == ord("1")  # shape (num_unique, num_qubits)
    probabilities = counts / counts.sum()
    return config_matrix, probabilities, permuted_bitstrings


config_matrix, probabilities, bitstrings = load_sqdrift_samples(
    "shots/sqdrift/001/sqdrift_circuits__id_001__time_1.0__ngroups_15_randomization_0_shots.npz",
    "shots/sqdrift_permutations/001/sqdrift_circuits__id_001__time_1.0__ngroups_15_randomization_0_metadata.txt",
)
Loading every circuit for a molecule

The helper below globs every *_shots.npz file in a molecule's SqDRIFT samples folder, loads and permutes each one via load_sqdrift_samples, and merges them into a single configuration matrix with summed counts (duplicate configurations, including ones only distinguished before permutation, collapse into one row with the combined count). This pools samples across randomizations and time points for one molecule into a single sample set.

def load_sqdrift_molecule(samples_dir, permutations_dir):
    """Load and merge all SqDRIFT samples for one molecule.

    `samples_dir` is e.g. "shots/sqdrift/001" (or ".../001/<config>" for the
    additional_sqdrift_configs_for_subset layout); `permutations_dir` is the
    matching "shots/sqdrift_permutations/001[/<config>]" folder.
    """
    all_bitstrings = []
    all_counts = []
    num_qubits = None

    for sample_path in sorted(glob.glob(os.path.join(samples_dir, "*_shots.npz"))):
        stem = re.sub(r"_shots$", "", os.path.splitext(os.path.basename(sample_path))[0])
        permutation_path = os.path.join(permutations_dir, f"{stem}_metadata.txt")

        config_matrix, _, bitstrings = load_sqdrift_samples(sample_path, permutation_path)
        with np.load(sample_path) as data:
            counts = data["count"].astype(np.int64)

        if num_qubits is None:
            num_qubits = config_matrix.shape[1]
        all_bitstrings.append(bitstrings)
        all_counts.append(counts)

    bitstrings = np.concatenate(all_bitstrings)
    counts = np.concatenate(all_counts)

    # Merge duplicate configurations (post-permutation) by summing their counts.
    unique_bitstrings, inverse = np.unique(bitstrings, return_inverse=True)
    merged_counts = np.zeros(len(unique_bitstrings), dtype=np.int64)
    np.add.at(merged_counts, inverse, counts)

    config_matrix = np.array(
        [[c == "1" for c in bs] for bs in unique_bitstrings], dtype=bool
    )
    probabilities = merged_counts / merged_counts.sum()
    return config_matrix, probabilities, unique_bitstrings


config_matrix, probabilities, bitstrings = load_sqdrift_molecule(
    "shots/sqdrift/001", "shots/sqdrift_permutations/001"
)

Files

additional_sqdrift_configs_for_subset_inventory.txt

Files (34.4 GB)

Name Size
md5:0fe54634beca0cea97d350466a3ca8c0
227.4 MB Preview Download
md5:fd08fdc66d2198375b51f6b5f1fe8464
524.3 MB Download
md5:64206c944676d3ce9bc8c947aea7f724
524.3 MB Download
md5:9ae16e31bc9e5f1393dfff017104f49c
524.3 MB Download
md5:b4d3eea8794a5758aea8913eef581b5d
524.3 MB Download
md5:cb016fa52ca280f707046277cdc84a79
524.3 MB Download
md5:b183bd54d27832385fe731ea92457e73
524.3 MB Download
md5:271f491ab56ca27133f1debba78d904a
524.3 MB Download
md5:6e5bbb53215feadbe6bb20f71cef3d44
524.3 MB Download
md5:7f7e07f3f75823c7724c40ab9afaa17d
524.3 MB Download
md5:ec24d31f479e717197663339b7eec35d
524.3 MB Download
md5:ff2370e41bd19c34f2e51136aa3d21b5
524.3 MB Download
md5:89ad0a40150771614af624185bd1ee22
524.3 MB Download
md5:38f4eb8bbffae27390734237cd9f9459
208.4 MB Download
md5:6a284ed480d603707d6d42f4bb3995cd
33.7 MB Download
md5:106349beda1eee43e0b60f11ee26ada5
524.3 MB Download
md5:01aa0533ea16810dba8634be7ef4a07f
524.3 MB Download
md5:d8a93d291fd284b502d12292f0b42181
524.3 MB Download
md5:10c3385889e5162793c4ea9eda945bfc
512.5 MB Download
md5:c026f65377ff3547aa1b96a32341bc29
229.0 MB Preview Download
md5:c9fd4fc0f8d1507270a46df37af8e8e4
2.4 kB Preview Download
md5:ac9eb2200b227d82a3ed5595429915ee
524.3 MB Download
md5:10e0e79b41aaa8d133ccc013450e7271
524.3 MB Download
md5:40a8c0460beb30125b6410549b61977c
524.3 MB Download
md5:5d93b2d6df9b6296a64b65abbd67901e
524.3 MB Download
md5:fb0b7c4777c4c1f169533d2dd6596830
524.3 MB Download
md5:bf09f9c9dc5ce680f261293b3ba69ead
524.3 MB Download
md5:52d87d7ff18cd20deb8164c8d0e36a23
524.3 MB Download
md5:6d82a695f8961bd5947a01329d446cf7
524.3 MB Download
md5:656db1812cc6b65a531fadff6764a20d
524.3 MB Download
md5:a14f1f2d4538cab549ef7f65fb2cf874
524.3 MB Download
md5:014d1d0b1f92895bb5bdb0371d66e19b
524.3 MB Download
md5:7b5f0103263ef60aa248e3703ee6c115
524.3 MB Download
md5:7539fc44cda269862b9b99fab8b7b0c8
524.3 MB Download
md5:b357ebdda52e0ca07d78e9a22574eda4
524.3 MB Download
md5:60ea1ab59881558854c109cb6ed3c3ef
524.3 MB Download
md5:58029478312dd6a390e76b8709d99af7
524.3 MB Download
md5:82a2632ee507a9c6bf2b4fc5c0953289
524.3 MB Download
md5:4e21097fb70600f0514dcb316a0a2e97
524.3 MB Download
md5:26c33f1abaf6bd014e7a004a07a74971
524.3 MB Download
md5:a33d7d4ef3bafe7ca5393977fd00563d
524.3 MB Download
md5:15ec0cdddcf630eb73e478c7395db8d9
524.3 MB Download
md5:cf1377d37e97eb4bfed54b7d8d717758
524.3 MB Download
md5:ef343fc8940cfb0faa4cb1e8a5090fec
524.3 MB Download
md5:4d91091b4f21d870224ebd24e1959ab2
524.3 MB Download
md5:4743a029c4d106977c519f576b2edaaa
84.8 MB Download
md5:5e49d2196e8f77c90bb616b473a8cb5d
57.8 kB Preview Download
md5:fb4508916d8841dc17534c075c80c0dc
12.2 kB Preview Download
md5:c0ff86bb60287376b3e87bd806623fa5
524.3 MB Download
md5:66af134158b19ff7b30895ab1e54edf2
524.3 MB Download
md5:877bfa87493ae53c888685074b254c7a
524.3 MB Download
md5:febf003b8aaf20ea0a3a731533a9241e
524.3 MB Download
md5:22e5e7bdac257494a19bab846258d112
524.3 MB Download
md5:dd4b648016fc09a558fafbfef0b35771
524.3 MB Download
md5:0a274f7200b1fc6ebf8efca73fb2b59f
524.3 MB Download
md5:57aa4bbdd33c0e21d02d29ff6266f1ea
524.3 MB Download
md5:56f78c16f87654c4456db03741b6e80f
524.3 MB Download
md5:c5daa25c00e0511a186d377a911305f2
524.3 MB Download
md5:03205abdb94fd44d28345725dbd828e0
524.3 MB Download
md5:cb89d4bf995efe0b29bc525dc3e5092d
524.3 MB Download
md5:acf02b9a56a69bc2a1df39e7bd8a7cf7
524.3 MB Download
md5:8c0488183e0d15e8414e0a1affa623ee
524.3 MB Download
md5:cb4d97be7eb9ba596322c5816124e4c9
524.3 MB Download
md5:df4fbd23c3672ac7237d9194925aa16c
524.3 MB Download
md5:9a4fd10c01f0ab2a112764ff800534fa
524.3 MB Download
md5:32b10618eb3b749fb6712662a28a95b8
524.3 MB Download
md5:850cb010a1ab05bc199995f884b3bbfe
524.3 MB Download
md5:0e81dcf2a2aa963157a984ca43369336
524.3 MB Download
md5:ddd67b7684a25dfd3512aaa886145d42
524.3 MB Download
md5:82234b19636d7726cd2cb002cfd589ea
524.3 MB Download
md5:81a336407012fd07d485e83780cdf3c4
524.3 MB Download
md5:11fbd501c23601a33709e9dec5852bee
524.3 MB Download
md5:d8f2ebd69d5c20f56711a0527723f2ae
75.7 MB Download
md5:7a971b6966c91b54b2f56e8c2368d334
46.3 MB Download