Code and Data for Chernoff-Based Throughput Optimization in Superconducting Qubit Readout

# Chernoff-based throughput optimization in superconducting-qubit readout Code and data for *Beyond single-shot fidelity: Chernoff-based throughputoptimization in superconducting-qubit readout* (S. Bugu). `Paper_MASTER_ALL_20260731.py` is the only script. It reproduces every figureand every numerical value in the manuscript and Supplemental Material. ## Requirements Python 3.10+, `numpy`, `scipy`, `pandas`, `matplotlib`.Developed on Python 3.12 with numpy 1.26.4, scipy 1.11.4, pandas 2.1.4,matplotlib 3.8.2. A shim maps `np.trapezoid` to `np.trapz` on numpy < 2.0, soboth numpy generations work. ## Running python3 Paper_MASTER_ALL_20260724.py --stage all Stages: | `--stage` | Work done ||--------------|------------------------------------------------------------------|| `finite` | P4/P4b exact finite-shot sweep; writes the three primary CSVs || `main` | Figures 1-6 || `supplement` | Figures S1-S3, Tables S1-S5 data, `supplement_tables.csv`, `fast_reset_scan.csv`, `audit.txt` || `all` | all of the above (default) | `--force` deletes the finite-shot cache and recomputes it. The finite-shot sweep dominates the runtime (order hours on one core for thefull twenty points). Results are cached per point as JSON, so reruns of `main`and `supplement` are fast. Deleting `FiniteShot_Cache/` forces a fullrecomputation; deleting only `FiniteShot_Cache/p4b/` reuses the P4 stage andrecomputes the refinement, which preserves `tau_fid`, `tau_C`, `G` and theBahadur-Rao presets exactly. ## Generated files Directories are created on first run. ### `Generated_Data/` | File | Contents ||-----------------------------------|------------------------------------------------------------------------------------------------------------|| `p4b_summary.csv` | One row per parameter point (20). Asymptotic optima, convergence diagnostics, and regret statistics for the four fixed rules. Source of Tables S1-S4. || `supplement_tables.csv` | Tables S1-S4 merged, in the units and rounding printed in the paper. 20 rows x 37 columns. || `fig6_finite_shot_regret_data.csv`| Regret curves and target-specific optima at both baselines. Source of Fig. 6 and Fig. S1. || `br_plateau_data.csv` | Bahadur-Rao prefactor diagnostic. Source of Fig. S3. | | `fast_reset_scan.csv` | Standard-linewidth total-overhead scan from 0 to 30 us. Source of Table S5. || `audit.txt` | Every count and range quoted in the manuscript, re-derived from `p4b_summary.csv`, plus three invariant checks. | ### `Figures/` `fig1_physical_model.pdf` ... `fig6_finite_shot_calibration.pdf`, with theplotted data alongside as `fig1a_data.csv`, `fig1b_data.csv`, `fig1c_data.csv`,`fig2_scan_data.csv`, `fig3_data.csv`, `fig4a_heatmap_data.csv`,`fig4b_heatmap_data.csv`, `fig5a_data.csv`, `fig5b_data.csv`. ### `SI_Figures/` `figS1_finite_shot_calibration_submhz.pdf`, `figS2_rule_comparison.pdf`,`figS3_br_plateau.pdf` (PDF and PNG), with `figS1_data.csv`, `figS2_data.csv`,`figS3_data.csv`. ## Model and survey Two linewidth regimes, ten parameter points each. - `standard`: chi/2pi = 1.2 MHz, kappa/2pi = 5.0 MHz (main-text baseline)- `manuscript`: chi/2pi = 0.19 MHz, kappa/2pi = 0.80 MHz (Supplemental Material) Points vary T1 (20-140 us), eta (0.30-0.65), nbar (40-120) and tau_oh(5-30 us) around T1 = 30 us, eta = 0.45, nbar = 80, tau_oh = 15 us. Theinternal name `P07_ceiling` is displayed as `P07_highNbar`; internal names arefrozen so old caches stay readable. Numerics: 4000-point score grid, 1500-point time grid, 80-node midpointquadrature over jump times, 16384-bin log-likelihood-ratio grid. Exact N-shoterror probabilities are computed by exponentially tilted FFT convolution andagree with the closed-form Gaussian result to 6e-5 relative down toP_e ~ 2e-54. Calibration regret is evaluated over log10(1/eps) in [3, 12]. ## Contact Sinan Bugu, [e-mail]