diff --git a/python/lammps/numpy_wrapper.py b/python/lammps/numpy_wrapper.py
index f3ea0fdf8e..a29853d16a 100644
--- a/python/lammps/numpy_wrapper.py
+++ b/python/lammps/numpy_wrapper.py
@@ -63,12 +63,16 @@ class numpy_wrapper:
 
     .. note::
 
-       While the returned arrays of per-atom data are dimensioned
-       for the range [0:nmax] - as is the underlying storage -
-       the data is usually only valid for the range of [0:nlocal],
-       unless the property of interest is also updated for ghost
-       atoms.  In some cases, this depends on a LAMMPS setting, see
-       for example :doc:`comm_modify vel yes <comm_modify>`.
+       The returned arrays of per-atom data are by default dimensioned
+       for the range [0:nlocal] since that data is *always* valid.  The
+       underlying storage for the data, however, is typically allocated
+       for the range of [0:nmax]. Whether there is valid data in the range
+       [nlocal:nlocal+nghost] depends on whether the property of interest
+       is also updated for ghost atoms.  This is not often the case.  In
+       some cases, it depends on a LAMMPS setting, see for example
+       :doc:`comm_modify vel yes <comm_modify>`.  By using the optional
+       *nelem* parameter the size of the returned NumPy can be overridden.
+       There is no check whether the number of elements chosen is valid.
 
     :param name: name of the property
     :type name:  string