Fix left multiplying operators with numpy arrays (#5361)

[sc-58675]

**Context:**

`np.array([0.5, 0.6]) * qml.S(0)` should give a `SProd` op with a batch
dimension.

**Description of the Change:**

Setting the `__array_priority__` property tells numpy we should use
`Operator.__mul__` and `Operator.__rmul__` instead of the corresponding
numpy methods that try and stick an operator into an array.

**Benefits:**

**Possible Drawbacks:**

**Related GitHub Issues:**

doc/releases/changelog-dev.md

@@ -60,6 +60,9 @@
 
 <h3>Improvements 🛠</h3>
 
+* Operators can now be left multiplied `x * op` by numpy arrays.
+  [(#5361)](https://github.com/PennyLaneAI/pennylane/pull/5361)
+
 * Create the `qml.Reflection` operator, useful for amplitude amplification and its variants.
   [(##5159)](https://github.com/PennyLaneAI/pennylane/pull/5159)
 

pennylane/operation.py

@@ -691,6 +691,10 @@ def compute_decomposition(theta, wires):
 
     # pylint: disable=too-many-public-methods, too-many-instance-attributes
 
+    # this allows scalar multiplication from left with numpy arrays np.array(0.5) * ps1
+    # taken from [stackexchange](https://stackoverflow.com/questions/40694380/forcing-multiplication-to-use-rmul-instead-of-numpy-array-mul-or-byp/44634634#44634634)
+    __array_priority__ = 1000
+
     def __init_subclass__(cls, **_):
         register_pytree(cls, cls._flatten, cls._unflatten)
 

tests/test_operation.py

@@ -1197,6 +1197,24 @@ def test_mul_scalar_tensor(self):
         assert isinstance(prod_op, SProd)
         assert prod_op.scalar is scalar
 
+        prod_op2 = scalar * qml.RX(1.23, 0)
+        assert isinstance(prod_op2, SProd)
+        assert prod_op.scalar is scalar
+
+    def test_mul_array_numpy(self):
+        """Test that the __mul__ dunder works with a batched scalar."""
+
+        scalar = np.array([0.5, 0.6, 0.7])
+        prod_op = scalar * qml.S(0)
+        assert isinstance(prod_op, SProd)
+        assert prod_op.scalar is scalar
+        assert prod_op.batch_size == 3
+
+        prod_op = qml.S(0) * scalar
+        assert isinstance(prod_op, SProd)
+        assert prod_op.scalar is scalar
+        assert prod_op.batch_size == 3
+
     def test_divide_with_scalar(self):
         """Test the __truediv__ dunder method with a scalar value."""
         sprod_op = qml.RX(1, 0) / 4
@@ -1255,6 +1273,10 @@ def test_mul_scalar_torch_tensor(self):
         assert isinstance(prod_op, SProd)
         assert prod_op.scalar is scalar
 
+        prod_op = scalar * qml.RX(1.23, 0)
+        assert isinstance(prod_op, SProd)
+        assert prod_op.scalar is scalar
+
     @pytest.mark.tf
     def test_mul_scalar_tf_tensor(self):
         """Test the __mul__ dunder method with a scalar tf tensor."""
@@ -1265,6 +1287,10 @@ def test_mul_scalar_tf_tensor(self):
         assert isinstance(prod_op, SProd)
         assert prod_op.scalar is scalar
 
+        prod_op = scalar * qml.RX(1.23, 0)
+        assert isinstance(prod_op, SProd)
+        assert prod_op.scalar is scalar
+
     @pytest.mark.jax
     def test_mul_scalar_jax_tensor(self):
         """Test the __mul__ dunder method with a scalar jax tensor."""
@@ -1275,6 +1301,10 @@ def test_mul_scalar_jax_tensor(self):
         assert isinstance(prod_op, SProd)
         assert prod_op.scalar is scalar
 
+        prod_op = scalar * qml.RX(1.23, 0)
+        assert isinstance(prod_op, SProd)
+        assert prod_op.scalar is scalar
+
     def test_mul_with_operator(self):
         """Test the __matmul__ dunder method with an operator."""
         prod_op = qml.RX(1, 0) @ qml.PauliX(0)

tests/test_qubit_device.py

@@ -595,7 +595,7 @@ def test_non_analytic_expval(self, mock_qubit_device_with_original_statistics, m
             m.setattr("numpy.mean", lambda obs, axis=None: obs)
             res = dev.expval(obs)
 
-        assert res == obs
+        assert res == np.array(obs)  # no idea what is trying to cast obs to an array now.
 
     def test_no_eigval_error(self, mock_qubit_device_with_original_statistics):
         """Tests that an error is thrown if expval is called with an observable that does
@@ -675,7 +675,7 @@ def test_non_analytic_var(self, mock_qubit_device_with_original_statistics, monk
             m.setattr("numpy.var", lambda obs, axis=None: obs)
             res = dev.var(obs)
 
-        assert res == obs
+        assert res == np.array(obs)
 
     def test_no_eigval_error(self, mock_qubit_device_with_original_statistics):
         """Tests that an error is thrown if var is called with an observable that does not have eigenvalues defined."""

tests/test_qutrit_device.py

@@ -700,7 +700,7 @@ def test_non_analytic_expval(self, mock_qutrit_device_with_original_statistics,
             m.setattr("numpy.mean", lambda obs, axis=None: obs)
             res = dev.expval(obs)
 
-        assert res == obs
+        assert res == np.array(obs)
 
     def test_no_eigval_error(self, mock_qutrit_device_with_original_statistics):
         """Tests that an error is thrown if expval is called with an observable that does
@@ -762,7 +762,7 @@ def test_non_analytic_var(self, mock_qutrit_device_with_original_statistics, mon
             m.setattr("numpy.var", lambda obs, axis=None: obs)
             res = dev.var(obs)
 
-        assert res == obs
+        assert res == np.array(obs)
 
     def test_no_eigval_error(self, mock_qutrit_device_with_original_statistics):
         """Tests that an error is thrown if var is called with an observable that does not have eigenvalues defined."""