Add probability normalization to DynamicsBackend sampling routine

qiskit_dynamics/backend/backend_utils.py

@@ -146,21 +146,31 @@ def _get_memory_slot_probabilities(
 
 
 def _sample_probability_dict(
-    probability_dict: Dict, shots: int, seed: Optional[int] = None
+    probability_dict: Dict,
+    shots: int,
+    normalize_probabilities: bool = True,
+    seed: Optional[int] = None,
 ) -> List[str]:
     """Sample outcomes based on probability dictionary.
 
     Args:
         probability_dict: Dictionary representing probability distribution, with keys being
             outcomes, values being probabilities.
         shots: Number of shots.
+        normalize_probabilities: Whether or not to normalize the probabilities to sum to 1 before
+            sampling.
         seed: Seed to use in rng construction.
 
     Return:
         List: of entries of probability_dict, sampled according to the probabilities.
     """
     rng = np.random.default_rng(seed=seed)
     alphabet, probs = zip(*probability_dict.items())
+
+    if normalize_probabilities:
+        probs = np.array(probs)
+        probs = probs / probs.sum()
+
     return rng.choice(alphabet, size=shots, replace=True, p=probs)
 
 

qiskit_dynamics/backend/dynamics_backend.py

@@ -118,9 +118,9 @@ class DynamicsBackend(BackendV2):
       indicating that the ground state for the system Hamiltonian should be used, or an arbitrary
       ``Statevector`` or ``DensityMatrix``. Defaults to ``"ground_state"``.
     * ``normalize_states``: Boolean indicating whether to normalize states before computing outcome
-      probabilities. Defaults to ``True``. Setting to ``False`` can result in errors if the solution
-      tolerance results in probabilities with significant numerical deviation from a proper
-      probability distribution.
+      probabilities, and normalize probablities before sampling. Defaults to ``True``. Setting to
+      ``False`` can result in errors if the solution tolerance results in probabilities with
+      significant numerical deviation from a proper probability distribution.
     * ``meas_level``: Form of measurement output. Supported values are ``1`` and ``2``. ``1``
       returns IQ points and ``2`` returns counts. Defaults to ``meas_level == 2``.
     * ``meas_return``: Level of measurement data to return. For ``meas_level = 1`` ``"single"``
@@ -819,7 +819,10 @@ def default_experiment_result_function(
 
         # sample
         memory_samples = _sample_probability_dict(
-            memory_slot_probabilities, shots=backend.options.shots, seed=seed
+            memory_slot_probabilities,
+            shots=backend.options.shots,
+            normalize_probabilities=backend.options.normalize_states,
+            seed=seed,
         )
         counts = _get_counts_from_samples(memory_samples)
 

releasenotes/notes/normalize-probabilities-d729245bb3fe5f10.yaml

@@ -0,0 +1,5 @@
+---
+fixes:
+  - |
+    ``DynamicsBackend.options.normalize_states`` now also controls whether or not the probability
+    distribution over outcomes is normalized before sampling outcomes.