qem-grover-robustness

Grover's Algorithm robustness study under realistic noise with a PyTorch quantum error mitigation (QEM) pipeline and an interactive Next.js dashboard.

What is implemented

• Parameterized Grover circuit generation for n_qubits in [2, 8]
• Aer noise modeling with:
  • single- and two-qubit depolarizing errors
  • readout error
  • thermal relaxation (T1/T2 + gate times)
• Batch experiment utilities for:
  • success probability vs iteration count
  • success probability vs noise strength
• QEM supervised learning pipeline:
  • noisy/ideal paired dataset generation
  • PyTorch MLP and Autoencoder options
  • MAE and fidelity evaluation
• FastAPI backend endpoints for simulation, sweeps, training, and status
• Next.js dashboard with:
  • controls for qubits, marked state, shots, and noise knobs
  • Recharts analytics
  • React Three Fiber 3D probability field
  • real-time API inference with mitigated outputs
• Technical report draft in report/technical_report.md

Repository layout

• backend/src/quantum: circuits, noise models, simulator, metrics, dataset generation
• backend/src/ml: QEM models and training/inference service
• backend/src/api: FastAPI app and request/response schemas
• backend/scripts: batch experiment and model training scripts
• frontend/src/app: Next.js dashboard UI
• frontend/src/components/ProbabilityField.tsx: 3D quantum-state view
• report/technical_report.md: academic report draft

Backend setup

cd backend
python -m venv .venv
# Windows:
.venv\Scripts\activate
# macOS/Linux:
# source .venv/bin/activate
pip install -r requirements.txt
uvicorn src.api.main:app --reload --port 8000

Backend docs: http://localhost:8000/docs

Frontend setup

cd frontend
npm install
npm run dev

Dashboard: http://localhost:3000

Set NEXT_PUBLIC_API_URL if backend is not on http://localhost:8000.

Key API routes

• GET /: health check
• GET /api/grover/ideal: circuit metadata
• POST /api/grover/run: ideal/noisy/mitigated single run
• POST /api/grover/sweep/iterations: success vs iteration
• POST /api/grover/sweep/noise: success vs noise
• POST /api/qem/train: train QEM model for specific qubit count
• GET /api/qem/status: list loaded QEM models

Batch experiments

From repository root:

python backend/scripts/run_batch_experiments.py --shots 10000 --include-mitigation

Outputs CSV files under data/experiments/:

• iteration_sweep.csv
• noise_sweep.csv
• failure_thresholds.csv

QEM model training CLI

python backend/scripts/train_qem.py --n-qubits 4 --sample-count 512 --epochs 80

Model checkpoint is written to data/experiments/qem_model.pt by default.