Fix VQD for k>2 (#8989)

* Fix vqd for k>2

Also ensure the optimal circuits from previous iterations are only constructed once and recognizable by the primitive caches.

* Remove empty parameter list

Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
(cherry picked from commit aab18fc)

qiskit/algorithms/eigensolvers/vqd.py

@@ -91,13 +91,16 @@ class VQD(VariationalAlgorithm, Eigensolver):
             optimizer(Optimizer): A classical optimizer. Can either be a Qiskit optimizer or a callable
                 that takes an array as input and returns a Qiskit or SciPy optimization result.
             k (int): the number of eigenvalues to return. Returns the lowest k eigenvalues.
-            betas (list[float]): beta parameters in the VQD paper.
+            betas (list[float]): Beta parameters in the VQD paper.
                 Should have length k - 1, with k the number of excited states.
                 These hyper-parameters balance the contribution of each overlap term to the cost
                 function and have a default value computed as the mean square sum of the
                 coefficients of the observable.
+            initial point (list[float]): An optional initial point (i.e. initial parameter values)
+                for the optimizer. If ``None`` then VQD will look to the ansatz for a preferred
+                point and if not will simply compute a random one.
             callback (Callable[[int, np.ndarray, float, dict[str, Any]], None] | None):
-                a callback that can access the intermediate data
+                A callback that can access the intermediate data
                 during the optimization. Four parameter values are passed to the callback as
                 follows during each evaluation by the optimizer: the evaluation count,
                 the optimizer parameters for the ansatz, the estimated value, the estimation
@@ -124,16 +127,16 @@ def __init__(
             ansatz: A parameterized circuit used as ansatz for the wave function.
             optimizer: A classical optimizer. Can either be a Qiskit optimizer or a callable
                 that takes an array as input and returns a Qiskit or SciPy optimization result.
-            k: the number of eigenvalues to return. Returns the lowest k eigenvalues.
-            betas: beta parameters in the VQD paper.
+            k: The number of eigenvalues to return. Returns the lowest k eigenvalues.
+            betas: Beta parameters in the VQD paper.
                 Should have length k - 1, with k the number of excited states.
                 These hyperparameters balance the contribution of each overlap term to the cost
                 function and have a default value computed as the mean square sum of the
                 coefficients of the observable.
-            initial_point: an optional initial point (i.e. initial parameter values)
+            initial_point: An optional initial point (i.e. initial parameter values)
                 for the optimizer. If ``None`` then VQD will look to the ansatz for a preferred
                 point and if not will simply compute a random one.
-            callback: a callback that can access the intermediate data
+            callback: A callback that can access the intermediate data
                 during the optimization. Four parameter values are passed to the callback as
                 follows during each evaluation by the optimizer: the evaluation count,
                 the optimizer parameters for the ansatz, the estimated value,
@@ -238,11 +241,19 @@ def compute_eigenvalues(
         if aux_operators is not None:
             aux_values = []
 
+        # We keep a list of the bound circuits with optimal parameters, to avoid re-binding
+        # the same parameters to the ansatz if we do multiple steps
+        prev_states = []
+
         for step in range(1, self.k + 1):
 
+            # update list of optimal circuits
+            if step > 1:
+                prev_states.append(self.ansatz.bind_parameters(result.optimal_points[-1]))
+
             self._eval_count = 0
             energy_evaluation = self._get_evaluate_energy(
-                step, operator, betas, prev_states=result.optimal_parameters
+                step, operator, betas, prev_states=prev_states
             )
 
             start_time = time()
@@ -304,15 +315,15 @@ def _get_evaluate_energy(
         step: int,
         operator: BaseOperator | PauliSumOp,
         betas: Sequence[float],
-        prev_states: list[np.ndarray] | None = None,
+        prev_states: list[QuantumCircuit] | None = None,
     ) -> Callable[[np.ndarray], float | list[float]]:
         """Returns a function handle to evaluate the ansatz's energy for any given parameters.
             This is the objective function to be passed to the optimizer that is used for evaluation.
 
         Args:
             step: level of energy being calculated. 0 for ground, 1 for first excited state...
             operator: The operator whose energy to evaluate.
-            prev_states: List of parameters from previous rounds of optimization.
+            prev_states: List of optimal circuits from previous rounds of optimization.
 
         Returns:
             A callable that computes and returns the energy of the hamiltonian
@@ -336,10 +347,6 @@ def _get_evaluate_energy(
 
         self._check_operator_ansatz(operator)
 
-        prev_circs = []
-        for state in range(step - 1):
-            prev_circs.append(self.ansatz.bind_parameters(prev_states[state]))
-
         def evaluate_energy(parameters: np.ndarray) -> np.ndarray | float:
 
             try:
@@ -355,10 +362,9 @@ def evaluate_energy(parameters: np.ndarray) -> np.ndarray | float:
             if step > 1:
                 # Compute overlap cost
                 fidelity_job = self.fidelity.run(
-                    [self.ansatz] * len(prev_circs),
-                    prev_circs,
-                    [parameters] * len(prev_circs),
-                    [prev_states[:-1]],
+                    [self.ansatz] * (step - 1),
+                    prev_states,
+                    [parameters] * (step - 1),
                 )
                 costs = fidelity_job.result().fidelities
 

releasenotes/notes/fix-vqd-kgt2-1ed95de3e32102c1.yaml

@@ -0,0 +1,5 @@
+---
+fixes:
+  - |
+    Fixed the :class:`~.eigensolvers.VQD` if more than ``k=2`` eigenvalues are computed.
+    Previously the code failed due to an internal type mismatch, but now runs as expected.

test/python/algorithms/eigensolvers/test_vqd.py

@@ -60,7 +60,7 @@ def setUp(self):
         algorithm_globals.random_seed = self.seed
 
         self.h2_energy = -1.85727503
-        self.h2_energy_excited = [-1.85727503, -1.24458455]
+        self.h2_energy_excited = [-1.85727503, -1.24458455, -0.88272215, -0.22491125]
 
         self.ryrz_wavefunction = TwoLocal(
             rotation_blocks=["ry", "rz"], entanglement_blocks="cz", reps=1
@@ -88,7 +88,7 @@ def test_basic_operator(self, op):
 
         with self.subTest(msg="test eigenvalue"):
             np.testing.assert_array_almost_equal(
-                result.eigenvalues.real, self.h2_energy_excited, decimal=1
+                result.eigenvalues.real, self.h2_energy_excited[:2], decimal=1
             )
 
         with self.subTest(msg="test dimension of optimal point"):
@@ -108,6 +108,14 @@ def test_basic_operator(self, op):
             )
             np.testing.assert_array_almost_equal(job.result().values, result.eigenvalues, 6)
 
+    def test_full_spectrum(self):
+        """Test obtaining all eigenvalues."""
+        vqd = VQD(self.estimator, self.fidelity, self.ryrz_wavefunction, optimizer=L_BFGS_B(), k=4)
+        result = vqd.compute_eigenvalues(H2_PAULI)
+        np.testing.assert_array_almost_equal(
+            result.eigenvalues.real, self.h2_energy_excited, decimal=2
+        )
+
     @data(H2_PAULI, H2_OP)
     def test_mismatching_num_qubits(self, op):
         """Ensuring circuit and operator mismatch is caught"""
@@ -198,7 +206,7 @@ def test_vqd_optimizer(self, op):
         def run_check():
             result = vqd.compute_eigenvalues(operator=op)
             np.testing.assert_array_almost_equal(
-                result.eigenvalues.real, self.h2_energy_excited, decimal=3
+                result.eigenvalues.real, self.h2_energy_excited[:2], decimal=3
             )
 
         run_check()
@@ -226,7 +234,7 @@ def test_aux_operators_list(self, op):
         # Start with an empty list
         result = vqd.compute_eigenvalues(op, aux_operators=[])
         np.testing.assert_array_almost_equal(
-            result.eigenvalues.real, self.h2_energy_excited, decimal=2
+            result.eigenvalues.real, self.h2_energy_excited[:2], decimal=2
         )
         self.assertIsNone(result.aux_operators_evaluated)
 
@@ -236,7 +244,7 @@ def test_aux_operators_list(self, op):
         aux_ops = [aux_op1, aux_op2]
         result = vqd.compute_eigenvalues(op, aux_operators=aux_ops)
         np.testing.assert_array_almost_equal(
-            result.eigenvalues.real, self.h2_energy_excited, decimal=2
+            result.eigenvalues.real, self.h2_energy_excited[:2], decimal=2
         )
         self.assertEqual(len(result.aux_operators_evaluated), 2)
         # expectation values
@@ -250,7 +258,7 @@ def test_aux_operators_list(self, op):
         extra_ops = [*aux_ops, None, 0]
         result = vqd.compute_eigenvalues(op, aux_operators=extra_ops)
         np.testing.assert_array_almost_equal(
-            result.eigenvalues.real, self.h2_energy_excited, decimal=2
+            result.eigenvalues.real, self.h2_energy_excited[:2], decimal=2
         )
         self.assertEqual(len(result.aux_operators_evaluated), 2)
         # expectation values
@@ -278,7 +286,7 @@ def test_aux_operators_dict(self, op):
         # Start with an empty dictionary
         result = vqd.compute_eigenvalues(op, aux_operators={})
         np.testing.assert_array_almost_equal(
-            result.eigenvalues.real, self.h2_energy_excited, decimal=2
+            result.eigenvalues.real, self.h2_energy_excited[:2], decimal=2
         )
         self.assertIsNone(result.aux_operators_evaluated)