Bernstein-Vazirani algorithm

Bernstein-Vazirani experiments executed across ideal simulation, backend-noise emulation, custom noise injections, and optional IBM Quantum hardware, with automated plotting and metric aggregation.

Algorithm

• Implements the Bernstein-Vazirani oracle and circuit for 5-qubit secrets (one ancillary qubit for phase kickback).
• Evaluates a fixed suite of 25 secret bitstrings to benchmark robustness under varied noise conditions.
• Single-shot oracle query recovers the hidden bitstring; correctness is assessed via the most frequent measurement outcome.

Experiment flow

• Ideal simulation: Aer simulator without noise for each secret.
• Backend-noise emulation: Aer simulator with noise models derived from ibm_torino and ibm_fez when available.
• Real hardware (optional): runtime Sampler execution on ibm_torino and ibm_fez; missing backends are reported and skipped.
• Readout sweeps: injected symmetric readout errors across predefined levels (ultra-low to very-high) with ideal overlays.
• Custom noise sweeps: depolarizing, amplitude/phase damping, thermal relaxation, and gate-specific noise across the same intensity levels.
• Plots: horizontal histograms comparing ideal, emulated, noisy, and real results; readout and custom sweeps are saved per secret and noise kind.

Setup

• Create an isolated Python environment and install dependencies:
  • python -m venv .venv && source .venv/bin/activate
  • pip install -r requirements.txt
• Populate .env from .env.example when using cloud backends.

Environment variables

• QISKIT_IBM_TOKEN
  • IBM Quantum API token used by QiskitRuntimeService for authentication.
• QISKIT_IBM_INSTANCE
  • IBM Cloud instance CRN to scope backend access.
• QISKIT_IBM_CHANNEL or IBM_QUANTUM_SERVICE
  • Service channel passed to QiskitRuntimeService; if omitted, default channel resolution is used.

Running

• Execute the Bernstein-Vazirani suite:
  • python src/bv.py
• The script targets ibm_torino and ibm_fez for backend-noise emulation and real execution; other backends are ignored.

Outputs

• Predictions: Plot_results/bv_predictions.csv with per-run category, backend, expected secret, predicted bitstring, correctness flag, top outcome probability, and raw counts.
• Metrics: Plot_results/bv_metrics.csv with accuracy, macro precision/recall, and macro F1 per (category, backend) pair, including noise annotations.
• Histograms: PNG files under Plot_results/ grouped by category:
  • ideal/ and noisy/ histograms per secret and backend.
  • backend_comparison/ overlays of ideal vs emulated vs real (when available).
  • custom_noise/readout/ sweep plots per secret; analogous folders for each custom noise kind.
• Cached counts: JSON histograms under Plot_results/hist_data/ to avoid recomputation on subsequent runs.