Hi, so would the following work for coupling the impurity site to the rest of your system, assuming the impurity site is site number 1?
(1) remove the "U" term acting on site 1 when defining the AutoMPO
(2) remove the "t" hopping term connecting site 1 to any other sites (such as site 2 for a 1D chain)
(3) include the S.S interaction using the following lines:
ampo += J,"Sz",1,"Sz",2;
ampo += J/2.,"S+",1,"S-",2;
ampo += J/2.,"S-",1,"S+",2;
(If you're wondering about what is the set of operators that are defined for the Hubbard site set the complete list is here: http://itensor.org/docs.cgi?page=classes/hubbard )
Then when making the initial state (using the InitState feature), make sure there is exactly one fermion on site 1 (not zero or two fermions). The spin of this fermion shouldn't matter except for determining the overall or total Sz quantum number of your many-body wavefunction (since the dmrg code preserves the total quantum number of whatever state you use as your initial state).
Trying this out on a toy system right now where I set J=1 and all of the other couplings to zero, I get a ground state energy of -0.75 as expected for the ground state of a two-spin Heisenberg model.
If I misunderstood your question or if that doesn't work please let me know.