The commands send and recv are high-level commands. That means they do under-the-hood work that saves the programmer time and avoids possible errors. They allocate memory after having internally deduced the datatype and the amount of buffer data needed for communication. This is done internally on a lower level based on C constructions.
NumPy arrays are objects that themselves make use of these C-buffer-like objects, so when sending and receiving NumPy arrays you can gain efficiency by using them in the lower-level communication counterparts Send and Recv (mind the capitalization!).
In the following example, we send an array from one processor to another:
from mpi4py import MPI
comm=MPI.COMM_WORLD # making a communicator instance
rank=comm.Get_rank() # querying for the numeric identifier of the core
size=comm.Get_size() # the total number of cores assigned
import numpy as np
if rank==0:
A = np.arange(700)
comm.Send(A, dest=1...