Andrea's suggestions

src/qibo/backends/abstract.py

@@ -357,7 +357,14 @@ def calculate_matrix_power(
         self, matrix, power: Union[float, int]
     ):  # pragma: no cover
         """Calculates the (fractional) ``power`` :math:`\\alpha` of ``matrix`` :math:`A`,
-        i.e. :math:`A^{\\alpha}`."""
+        i.e. :math:`A^{\\alpha}`.
+        
+        .. note::
+            For the ``pytorch`` backend, this method relies on a copy of the original tensor.
+            This may break the gradient flow. For the GPU backends (i.e. ``cupy`` and 
+            ``cuquantum``), this method falls back to CPU whenever ``power`` is not
+            an integer.        
+        """
         raise_error(NotImplementedError)
 
     @abc.abstractmethod

src/qibo/quantum_info/linalg_operations.py

@@ -211,13 +211,6 @@ def matrix_power(matrix, power: Union[float, int], backend=None):
     Returns:
         ndarray: matrix power :math:`A^{\\alpha}`.
     """
-
-    if not isinstance(power, (float, int)):
-        raise_error(
-            TypeError,
-            f"``power`` must be either float or int, but it is type {type(power)}.",
-        )
-
     backend = _check_backend(backend)
 
     return backend.calculate_matrix_power(matrix, power)