[Bug] Fix GPSR differentiation mode (#616)

qadence/analog/parse_analog.py

@@ -78,9 +78,8 @@ def _build_rot_ham_evo(
     if block.add_pattern and h_addr is not None:
         h_block += h_addr
     duration = block.parameters.duration
-    h_norm = block.parameters.h_norm
     h_block += h_drive
-    return HamEvo(h_block / h_norm, duration * h_norm / 1000)
+    return HamEvo(h_block, duration / 1000)
 
 
 def _analog_to_hevo(

qadence/backends/gpsr.py

@@ -36,7 +36,11 @@ def general_psr(spectrum: Tensor, n_eqs: int | None = None, shift_prefac: float
     sorted_unique_spectral_gaps = torch.tensor(list(sorted_unique_spectral_gaps)[:n_eqs])
 
     if n_eqs == 1:
-        return single_gap_psr
+        return partial(
+            single_gap_psr,
+            spectral_gap=sorted_unique_spectral_gaps,
+            shift=shift_prefac * torch.tensor([PI / 2], dtype=torch.get_default_dtype()),
+        )
     else:
         return partial(
             multi_gap_psr,
@@ -110,7 +114,9 @@ def multi_gap_psr(
     batch_size = max(t.size(0) for t in param_dict.values())
 
     # get shift values
-    shifts = shift_prefac * torch.linspace(PI / 2 - PI / 5, PI / 2 + PI / 5, n_eqs)
+    shifts = shift_prefac * torch.linspace(
+        PI / 2 - PI / 4, PI / 2 + PI / 5, n_eqs
+    )  # breaking the symmetry of sampling range around PI/2
     device = torch.device("cpu")
     try:
         device = [v.device for v in param_dict.values()][0]

qadence/execution.py

@@ -248,8 +248,6 @@ def _(
     configuration: Union[BackendConfiguration, dict, None] = None,
 ) -> Tensor:
     observable = observable if isinstance(observable, list) else [observable]
-    if backend == BackendName.PYQTORCH:
-        diff_mode = DiffMode.AD
     bknd = backend_factory(backend, diff_mode=diff_mode, configuration=configuration)
     conv = bknd.convert(circuit, observable)
 

tests/backends/pulser_basic/test_differentiation.py

@@ -79,7 +79,7 @@ def test_pulser_gpsr(block_id: int) -> None:
     pulser_backend = PulserBackend()  # type: ignore[arg-type]
     conv = pulser_backend.convert(circ, obs)
     pulser_circ, pulser_obs, embedding_fn, params = conv
-    diff_backend = DifferentiableBackend(pulser_backend, diff_mode=DiffMode.GPSR, shift_prefac=0.2)
+    diff_backend = DifferentiableBackend(pulser_backend, diff_mode=DiffMode.GPSR, shift_prefac=1.0)
     expval_pulser = diff_backend.expectation(pulser_circ, pulser_obs, embedding_fn(params, values))
     dexpval_x_pulser = torch.autograd.grad(
         expval_pulser, values["x"], torch.ones_like(expval_pulser), create_graph=True