QAOA RBM Annealing Dashboard

Desktop dashboard and simulation workflow for studying QAOA-based annealing on RBM-inspired Ising models.

The project lets you sweep delta_t, compare estimated kappa against KL-divergence, and generate plots and result files for different visible/hidden-node configurations.

This repository contains two main usage paths:

• beta_kl_dashboard.py Opens a Tkinter desktop dashboard for interactive parameter sweeps and live plotting.

• QAOAAnnealing_Modified.py Runs the simulation directly from the command line for scripted experiments.

Supporting modules:

• plot_beta_vs_kl.py Loads saved results and exports a styled Kappa vs KL-Divergence figure.
• rbm_config.py Builds RBM parameter configurations shared across the workflow.
• rbm_kl_curve.py Generates the analytical KL-vs-kappa reference curve.

• Python 3
• Qiskit
• Qiskit Aer
• Qiskit Optimization
• NumPy
• SciPy
• Matplotlib

Install everything with:

python -m venv .venv
.\.venv\Scripts\activate
pip install -r requirements.txt

If you want a one-step launcher on Windows, use:

.\run_dashboard.ps1

Or the double-clickable wrapper:

.\run_dashboard.bat

.\.venv\Scripts\python.exe .\beta_kl_dashboard.py

The launcher is the safer option on a fresh checkout because it creates the virtual environment automatically when needed. If it reports that Tkinter could not initialize, repair or reinstall your local Python with Tcl/Tk support and rerun the launcher.

The dashboard supports:

• delta_t sweep ranges
• number of sweep points
• circuit depth and shots
• number of visible and hidden nodes
• annealing schedule selection
• Hamiltonian normalization selection
• optional sampled-count threshold filtering
• saved plots and JSON outputs for each run

Example:

.\.venv\Scripts\python.exe .\QAOAAnnealing_Modified.py --delta-t-values 0.01 0.02 0.03 --reps 100 --shots 1000000 --num-visible-nodes 7 --num-hidden-nodes 2 --hamiltonian-normalization largest-eigenvalue

Useful options include:

• --delta-t-values
• --delta-t-file
• --reps
• --shots
• --num-visible-nodes
• --num-hidden-nodes
• --annealing-schedule
• --enable-count-threshold-filter
• --rbm-seed
• --hamiltonian-normalization
• --kappa-estimator
• --execution-mode
• --aer-device

• linear Uses gamma(u) = u and beta(u) = 1 - u.
• dwave-like Uses the smooth D-Wave-inspired cubic schedule present in the project.
• linear-to-dwave Blends from the linear schedule toward the D-Wave-like schedule around the normalized midpoint, near kappa ~= 0.5.

• max-coefficient Divides the Ising Hamiltonian by the largest absolute Pauli coefficient.
• largest-eigenvalue Divides the Ising Hamiltonian by its spectral radius, max(|lambda_max|, |lambda_min|).

Typical runs can generate:

• results JSON files with delta_t, kappa_hat, kl_divergence, RBM sizes, seed, and normalization metadata
• a final Kappa vs KL-Divergence plot
• an optional minimum-KL diagnostic plot

• Hidden nodes refers to the size of a single hidden RBM layer.
• If --rbm-seed is omitted, a fresh random seed is generated and saved with the results.
• The analytical reference curve reuses the same saved RBM seed so the comparison stays consistent with the sampled run.
• On this Windows setup, Aer currently reports CPU-only availability.