---
license: mit
pretty_name: IBM 150-Qubit Ising Hardware Results
size_categories:
- n<1K
tags:
- quantum-computing
- qaoa
- ibm-quantum
- hardware-benchmark
- optimization
configs:
- config_name: depth_curve
  data_files:
  - split: train
    path: data/depth_curve.csv
- config_name: mitigation_ablation
  data_files:
  - split: train
    path: data/mitigation_ablation.csv
- config_name: replicated_confirmation
  data_files:
  - split: train
    path: data/replicated_confirmation.csv
- config_name: jobs
  data_files:
  - split: train
    path: data/jobs.csv
---

# IBM 150-Qubit Ising Hardware Results

Auditable results from a hardware-native, 150-active-qubit Ising/QAOA study on IBM `ibm_fez`, accompanied by an exact classical baseline and explicit scientific claim boundaries.

## Headline evidence

| Item | Verified value |
|---|---:|
| Active physical qubits | 150 |
| Processor qubits | 156 |
| Successful hardware jobs | 15 |
| QPU usage | 599 seconds |
| PUBs | 547 |
| Shots | 2,087,936 |
| QAOA depths | p=1 through p=8 |
| Exact classical optimum | 865 / 873, MIP gap 0 |
| Replicated best mean | p=6, 0.742274 |
| Replicated best sample | p=6, 0.915607 |

The 15 successful jobs consist of the 14 entries in the primary job ledger plus the final p5/p6 confirmation job. Together they reconcile exactly to 599 QPU seconds, 547 PUBs, and 2,087,936 shots.

## Configurations

- `depth_curve`: compiled circuit size and approximation ratios across QAOA depth p=1…8.
- `mitigation_ablation`: raw, dynamical-decoupling, randomized-twirling, and combined settings.
- `replicated_confirmation`: 22 independent replicates each for p=5 and p=6.
- `jobs`: the complete 15-job successful hardware ledger, including the final confirmation.

The `evidence/` directory retains the canonical release summary, experiment ledger, and source manifest. `SHA256SUMS` covers the packaged evidence and tables.

## Findings

The initial sweep reached its best mean approximation ratio at p=6 (0.739050). The independent confirmation again favored p=6 (0.742274 ± 0.000152 standard error). Dynamical decoupling alone slightly improved the p=1 mean, while twirling and the combined setting did not improve this workload.

## Truth boundary

- Hardware execution: verified.
- Exact classical baseline: verified.
- Quantum acceleration: not measured.
- Quantum advantage: not claimed.
- Production winner for this instance: classical HiGHS.

This is experimental evidence, not a claim that quantum optimization outperformed the exact classical solver.

## Provenance and reproducibility

- Canonical source: [sankalpsthakur/ibm-quantum-150q-ising](https://github.com/sankalpsthakur/ibm-quantum-150q-ising)
- Source commit: `7dd762389b7b45c8ab15f19f1508b4e0f96922cb`
- Experiment date: 2026-07-23
- Interactive visualization: [QAOA Depth & Noise Lab](https://huggingface.co/spaces/sankalpsthakur/qaoa-depth-noise-lab)

## Citation

```bibtex
@dataset{thakur_2026_ibm_150q_ising,
  author = {Sankalp Thakur},
  title = {IBM 150-Qubit Ising Hardware Results},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/sankalpsthakur/ibm-150q-ising-hardware-results}
}
```