symqnet-molopt 3.12.23

SymQNet Molecular Optimization via Hamiltonian Estimation

SymQNet-MolOpt: Hamiltonian Parameter Estimation

SymQNet-MolOpt provides efficient, uncertainty-aware estimation of Hamiltonian parameters for 1D and 2D molecular models and ultimately much more efficient molecular optimization. It is designed for sample-efficient optimization and reports confidence intervals for each parameter.

---

Installation

pip install SymQNet-MolOpt

---

Usage

Core Command

SymQNet-MolOpt --hamiltonian input.json --output results.json

Arguments:

• --hamiltonian: Path to a JSON Hamiltonian (OpenFermion-like format).
• --output: File to save results (JSON).
• --shots: Number of measurement shots (default auto-scales).
• --n-rollouts: Number of independent rollouts (default: 5).
• --max-steps: Max optimization steps per rollout (default: 50).

---

Examples

Water Molecule (H₂O, 10 qubits)

SymQNet-MolOpt --hamiltonian examples/H2O_10q.json --output h2o_results.json --shots 1024 --n-rollouts 5 --max-steps 50

Ising Chain (12 qubits)

SymQNet-MolOpt --hamiltonian examples/ising_12q.json --output ising_results.json --shots 1024

(You need to create your own JSON Hamiltonian.)

---

Input Format

Hamiltonians are specified in JSON:

{
  "format": "openfermion",
  "system": "H2O",
  "n_qubits": 10,
  "pauli_terms": [
    {"coefficient": -74.943, "pauli_string": "IIIIIIIIII"},
    {"coefficient": 0.342, "pauli_string": "IIIIIIIIIZ"}
  ]
}

---

Output Format

Results include estimated parameters with uncertainties plus experiment/config metadata:

{
  "symqnet_results": {
    "coupling_parameters": [
      {
        "index": 0,
        "mean": 0.2134,
        "confidence_interval": [0.2089, 0.2179],
        "uncertainty": 0.0045
      }
    ],
    "field_parameters": [...],
    "total_uncertainty": 0.0856,
    "avg_measurements_used": 512.0,
    "confidence_level": 0.95,
    "n_rollouts": 5
  },
  "hamiltonian_info": {
    "molecule": "H2O",
    "n_qubits": 10,
    "n_pauli_terms": 512,
    "format": "openfermion",
    "optimal_qubits": 10,
    "performance_optimal": true
  },
  "experimental_config": {
    "shots": 1024,
    "max_steps": 50,
    "n_rollouts": 5,
    "confidence": 0.95,
    "device": "cpu",
    "seed": 42
  },
  "metadata": {
    "generated_by": "Universal SymQNet Molecular Optimization CLI",
    "version": "2.0.12",
    "model_constraint": "Trained optimally for 10 qubits, supports any qubit count",
    "timestamp": "2024-01-01T00:00:00",
    "parameter_count": {
      "coupling": 9,
      "field": 10,
      "total": 19
    },
    "parameter_extraction_fixed": true
  },
  "performance_analysis": {
    "expected_performance": 1.0,
    "performance_level": "optimal",
    "optimal_qubits": 10,
    "universal_symqnet_version": "2.0.0"
  },
  "universal_wrapper": {
    "original_qubits": 12,
    "normalized_to": 10,
    "expected_performance": 0.85,
    "normalization_applied": true,
    "optimal_at": 10,
    "fixed_parameter_extraction": true
  },
  "validation": {
    "has_ground_truth": false
  }
}

---

Metadata Slots (Policy Input)

The policy network conditions on a metadata vector with fixed slots. The layout is defined in the architecture contract document: ARCHITECTURE_CONTRACT.md.

Slot layout (see MetadataLayout):

1. Qubit selection (one-hot): n_qubits slots.
2. Measurement basis (one-hot): 3 slots for X/Y/Z.
3. Evolution time (one-hot): M_evo slots.
4. Shot budget slot: single slot used only when shots_encoding is enabled in the checkpoint.
5. Posterior mean (θ): 2 * n_qubits - 1 slots.
6. Posterior covariance features: theta_dim + 8 slots.
7. Posterior Fisher diagonal: theta_dim slots.

Total metadata dimension is n_qubits + 3 + M_evo + 1 + theta_dim + (theta_dim + 8) + theta_dim.

---

Shot Handling

• The CLI --shots value sets the measurement simulator budget and is used by the SMC filter.
• If the checkpoint includes shots_encoding metadata, the shot budget is injected into the metadata vector at the shots_slot index using the normalized value: log1p(shots) / log1p(1_000_000). If shots_encoding is missing/null, the slot is left at 0.

---

Migration Note (Older Checkpoints)

Legacy checkpoints that do not include the metadata bundle will be rejected. Please re-export or retrain to include the required keys: model_state_dict, meta_dim, shots_encoding, n_qubits, M_evo, rollout_steps, plus the optional checkpoint_format/checkpoint_version fields. The current format expects checkpoint_format="symqnet-ppo-v2" and checkpoint_version=1.

---

Requirements

• Python 3.8+
• PyTorch 1.12+
• NumPy, SciPy, Click