Benchmark Summary

This page summarizes the benchmark coverage currently present under notebooks/benchmarks/ and the generated artifact state observed in the local workspace.

The benchmark notebooks are the source of truth. The ignored results/ and images/ directories are local/generated artifacts and are not packaged or committed.

Coverage At A Glance

Area

Notebooks

Main question answered

Cross-method chemistry comparisons

4

How do VQE, VarQITE, and QPE compare on H2 and LiH?

Multi-molecule chemistry panels

5

Which registry systems are ready for broader benchmarks, scaling, and ionization studies?

Default calibration

3

Are the package defaults justified by calibration sweeps?

Non-molecule model Hamiltonians

3

Does expert-mode Hamiltonian input work beyond chemistry?

QPE calibration and noise

3

How sensitive is QPE to ancilla count, evolution time, shots, and noise?

VQE H2 benchmark studies

10

How do ansatzes, mappings, noisy channels, and excited-state solvers behave on H2?

VQE H3+ benchmark studies

2

How do ansatzes behave on a larger hydrogen ion panel?

QITE / QRTE dynamics

1

How does projected real-time evolution compare with exact evolution on H2?

Total benchmark notebooks: 31.

Method Selection Notes

Need

Start with

Why

Fast ground-state baseline

VQE

Lowest overhead and broadest notebook coverage.

Adaptive ansatz construction

ADAPT-VQE

Useful when fixed ansatz choice is the main uncertainty.

Excited-state estimates after a reference VQE

QSE, EOM-QSE, LR-VQE, EOM-VQE

Reuses a converged VQE reference and exposes post-VQE excited-state diagnostics.

Direct variational excited states

SSVQE or VQD

Useful when excited states are the optimization target rather than a post-processing step.

Phase / spectral information

QPE

Best represented by the H2 calibration and decision-map notebooks.

Imaginary-time relaxation

VarQITE

Useful as an alternative state-preparation or relaxation path.

Real-time dynamics

VarQRTE

Covered by the exact QRTE benchmark on H2.

Non-chemistry qubit models

Expert-mode Hamiltonian inputs

Covered by TFIM, XXZ Heisenberg, and SSH benchmark notebooks.

Current Benchmark Notebook Inventory

Cross-Method Comparisons

  • comparisons/H2/Cross_Method_Comparison.ipynb

  • comparisons/H2/Reproducibility_Benchmark.ipynb

  • comparisons/LiH/Cross_Method_Comparison.ipynb

  • comparisons/LiH/Reproducibility_Benchmark.ipynb

Multi-Molecule Comparisons

  • comparisons/multi_molecule/Atomic_Ionization_Energy_Benchmark.ipynb

  • comparisons/multi_molecule/Hydrogen_Family_VQE_Benchmark.ipynb

  • comparisons/multi_molecule/Low_Qubit_VQE_Benchmark.ipynb

  • comparisons/multi_molecule/Registry_Coverage.ipynb

  • comparisons/multi_molecule/Scaling_Benchmark.ipynb

Default Calibration

  • defaults/QPE_Default_Calibration.ipynb

  • defaults/VQE_Default_Calibration.ipynb

  • defaults/VarQITE_Default_Calibration.ipynb

Non-Molecule Model Hamiltonians

  • non_molecule/Heisenberg_Chain_Benchmark.ipynb

  • non_molecule/SSH_Chain_Benchmark.ipynb

  • non_molecule/TFIM_Cross_Method_Benchmark.ipynb

QITE / QRTE

  • qite/H2/Exact_QRTE_Benchmark.ipynb

QPE

  • qpe/H2/Calibration_Decision_Map.ipynb

  • qpe/H2/Calibration_Sweep.ipynb

  • qpe/H2/Noisy.ipynb

VQE

  • vqe/H2/Ansatz_Comparison.ipynb

  • vqe/H2/Mapping_Comparison.ipynb

  • vqe/H2/Noise_Robustness_Benchmark.ipynb

  • vqe/H2/Noise_Scan.ipynb

  • vqe/H2/Noisy_Ansatz_Comparison.ipynb

  • vqe/H2/Noisy_Ansatz_Convergence.ipynb

  • vqe/H2/Noisy_Optimizer_Comparison.ipynb

  • vqe/H2/Noisy_Optimizer_Convergence.ipynb

  • vqe/H2/SSVQE_Comparisons.ipynb

  • vqe/H2/VQD_Comparisons.ipynb

  • vqe/H3plus/Ansatz_Comparison_Noiseless.ipynb

  • vqe/H3plus/Ansatz_Comparison_Noisy.ipynb

Artifact Audit

Observed local generated artifacts:

Directory

Files

Size

Git status

results/

733 JSON files

6.2 MiB

ignored

images/

3 PNG files

464 KiB

ignored

The result cache is useful for local reruns, but it is not clean enough to act as a committed benchmark dataset. It still mixes:

  • benchmark-style H2, H3+, H4, LiH, and hydrogen-family runs

  • QPE calibration sweeps with hundreds of per-seed/per-configuration JSON files

  • multiple historical cache signatures for similar benchmark configurations

Those generated cache artifacts are expected for local development, but they should not be interpreted as a curated benchmark dataset.

The stale smoke/cache result files and orphaned images identified during the 0.3.23 artifact audit have been removed from the local workspace.

Referenced generated images remaining locally:

  • images/vqe/H2/ansatz_conv_Adam_s0.png

  • images/vqe/H2/mapping_comparison_UCCSD_Adam.png

  • images/vqe/multi_molecule/low_qubit_benchmark_UCCSD_Adam_jordan_wigner_max10q.png