diff --git a/src/quemb/shared/typing.py b/src/quemb/shared/typing.py
index 643eeb97..8457420d 100644
--- a/src/quemb/shared/typing.py
+++ b/src/quemb/shared/typing.py
@@ -11,12 +11,18 @@
 
 import numpy as np
 
-T_co = TypeVar("T_co", bound=np.generic, covariant=True)
+# We want the dtype to behave covariant, i.e. if a
+#  Vector[float] is allowed, then the more specific
+#  Vector[float64] should also be allowed.
+#
+# Also see here:
+# https://stackoverflow.com/questions/61568462/what-does-typevara-b-covariant-true-mean
+T_dtype_co = TypeVar("T_dtype_co", bound=np.generic, covariant=True)
 
-Vector = np.ndarray[Tuple[int], np.dtype[T_co]]
-Matrix = np.ndarray[Tuple[int, int], np.dtype[T_co]]
-Tensor3D = np.ndarray[Tuple[int, int, int], np.dtype[T_co]]
-Tensor4D = np.ndarray[Tuple[int, int, int, int], np.dtype[T_co]]
-Tensor5D = np.ndarray[Tuple[int, int, int, int, int], np.dtype[T_co]]
-Tensor6D = np.ndarray[Tuple[int, int, int, int, int, int], np.dtype[T_co]]
-Tensor = np.ndarray[Tuple[int, ...], np.dtype[T_co]]
+Vector = np.ndarray[Tuple[int], np.dtype[T_dtype_co]]
+Matrix = np.ndarray[Tuple[int, int], np.dtype[T_dtype_co]]
+Tensor3D = np.ndarray[Tuple[int, int, int], np.dtype[T_dtype_co]]
+Tensor4D = np.ndarray[Tuple[int, int, int, int], np.dtype[T_dtype_co]]
+Tensor5D = np.ndarray[Tuple[int, int, int, int, int], np.dtype[T_dtype_co]]
+Tensor6D = np.ndarray[Tuple[int, int, int, int, int, int], np.dtype[T_dtype_co]]
+Tensor = np.ndarray[Tuple[int, ...], np.dtype[T_dtype_co]]