Fix zero-jvps with shot vectors

doc/releases/changelog-dev.md

@@ -53,6 +53,9 @@
 
 <h3>Bug fixes 🐛</h3>
 
+* Fix a bug where zero-valued JVPs were calculated wrongly in the presence of shot vectors.
+  [(#6219)](https://github.com/PennyLaneAI/pennylane/pull/6219)
+
 * Fix `qml.PrepSelPrep` template to work with `torch`:
   [(#6191)](https://github.com/PennyLaneAI/pennylane/pull/6191)
 
@@ -71,4 +74,5 @@ Utkarsh Azad,
 Lillian M. A. Frederiksen,
 Christina Lee,
 William Maxwell,
-Lee J. O'Riordan,
+Lee J. O'Riordan,
+David Wierichs,

pennylane/gradients/jvp.py

@@ -295,11 +295,16 @@ def jvp(tape, tangent, gradient_fn, gradient_kwargs=None):
     if len(tape.trainable_params) == 0:
         # The tape has no trainable parameters; the JVP
         # is simply none.
-        def zero_vjp(_):
-            res = tuple(np.zeros(mp.shape(None, tape.shots)) for mp in tape.measurements)
+        def zero_jvp_for_single_shots(s):
+            res = tuple(np.zeros(mp.shape(shots=s), dtype=mp.numeric_type) for mp in tape.measurements)
             return res[0] if len(tape.measurements) == 1 else res
 
-        return tuple(), zero_vjp
+        def zero_jvp(_):
+            if tape.shots.has_partitioned_shots:
+                return tuple(zero_jvp_for_single_shots(s) for s in tape.shots)
+            return zero_jvp_for_single_shots(tape.shots.total_shots)
+
+        return tuple(), zero_jvp
 
     multi_m = len(tape.measurements) > 1
 

pennylane/workflow/jacobian_products.py

@@ -46,6 +46,17 @@ def _compute_vjps(jacs, dys, tapes):
     return tuple(vjps)
 
 
+def _zero_jvp_single_shots(shots, tape):
+    jvp = tuple(np.zeros(mp.shape(shots=shots), dtype=mp.numeric_type) for mp in tape.measurements)
+    return jvp[0] if len(tape.measurements) == 1 else jvp
+
+
+def _zero_jvp(tape):
+    if tape.shots.has_partitioned_shots:
+        return tuple(_zero_jvp_single_shots(s, tape) for s in tape.shots)
+    return _zero_jvp_single_shots(tape.shots.total_shots, tape)
+
+
 def _compute_jvps(jacs, tangents, tapes):
     """Compute the jvps of multiple tapes, directly for a Jacobian and tangents."""
     f = {True: qml.gradients.compute_jvp_multi, False: qml.gradients.compute_jvp_single}
@@ -54,16 +65,7 @@ def _compute_jvps(jacs, tangents, tapes):
     for jac, dx, t in zip(jacs, tangents, tapes):
         multi = len(t.measurements) > 1
         if len(t.trainable_params) == 0:
-            empty_shots = qml.measurements.Shots(None)
-            zeros_jvp = tuple(
-                np.zeros(mp.shape(None, empty_shots), dtype=mp.numeric_type)
-                for mp in t.measurements
-            )
-            zeros_jvp = zeros_jvp[0] if len(t.measurements) == 1 else zeros_jvp
-            if t.shots.has_partitioned_shots:
-                jvps.append(tuple(zeros_jvp for _ in range(t.shots.num_copies)))
-            else:
-                jvps.append(zeros_jvp)
+            jvps.append(_zero_jvp(t))
         elif t.shots.has_partitioned_shots:
             jvps.append(tuple(f[multi](dx, j) for j in jac))
         else:

tests/gradients/core/test_jvp.py

@@ -17,6 +17,7 @@
 import pennylane as qml
 from pennylane import numpy as np
 from pennylane.gradients import param_shift
+from pennylane.measurements.shots import Shots
 
 _x = np.arange(12).reshape((2, 3, 2))
 
@@ -799,7 +800,8 @@ def cost_fn(params, tangent):
 class TestBatchJVP:
     """Tests for the batch JVP function"""
 
-    def test_one_tape_no_trainable_parameters(self):
+    @pytest.mark.parametrize("shots", [Shots(None), Shots(10), Shots([20, 10])])
+    def test_one_tape_no_trainable_parameters(self, shots):
         """A tape with no trainable parameters will simply return None"""
         dev = qml.device("default.qubit", wires=2)
 
@@ -808,46 +810,47 @@ def test_one_tape_no_trainable_parameters(self):
             qml.CNOT(wires=[0, 1])
             qml.expval(qml.PauliZ(0))
 
-        tape1 = qml.tape.QuantumScript.from_queue(q1)
+        tape1 = qml.tape.QuantumScript.from_queue(q1, shots=shots)
         with qml.queuing.AnnotatedQueue() as q2:
             qml.RX(0.4, wires=0)
             qml.RX(0.6, wires=0)
             qml.CNOT(wires=[0, 1])
             qml.expval(qml.PauliZ(0))
 
-        tape2 = qml.tape.QuantumScript.from_queue(q2)
+        tape2 = qml.tape.QuantumScript.from_queue(q2, shots=shots)
         tape1.trainable_params = {}
         tape2.trainable_params = {0, 1}
 
         tapes = [tape1, tape2]
         tangents = [np.array([1.0, 1.0]), np.array([1.0, 1.0])]
 
         v_tapes, fn = qml.gradients.batch_jvp(tapes, tangents, param_shift)
-        assert len(v_tapes) == 4
 
         # Even though there are 3 parameters, only two contribute
         # to the JVP, so only 2*2=4 quantum evals
+        assert len(v_tapes) == 4
         res = fn(dev.execute(v_tapes))
 
         assert qml.math.allclose(res[0], np.array(0.0))
         assert res[1] is not None
 
-    def test_all_tapes_no_trainable_parameters(self):
+    @pytest.mark.parametrize("shots", [Shots(None), Shots(10), Shots([20, 10])])
+    def test_all_tapes_no_trainable_parameters(self, shots):
         """If all tapes have no trainable parameters all outputs will be None"""
 
         with qml.queuing.AnnotatedQueue() as q1:
             qml.RX(0.4, wires=0)
             qml.CNOT(wires=[0, 1])
             qml.expval(qml.PauliZ(0))
 
-        tape1 = qml.tape.QuantumScript.from_queue(q1)
+        tape1 = qml.tape.QuantumScript.from_queue(q1, shots=shots)
         with qml.queuing.AnnotatedQueue() as q2:
             qml.RX(0.4, wires=0)
             qml.RX(0.6, wires=0)
             qml.CNOT(wires=[0, 1])
             qml.expval(qml.PauliZ(0))
 
-        tape2 = qml.tape.QuantumScript.from_queue(q2)
+        tape2 = qml.tape.QuantumScript.from_queue(q2, shots=shots)
         tape1.trainable_params = set()
         tape2.trainable_params = set()