1) To what "published value" did you compare your results for the silver grain boundary? To the experimental value, to another simulation study for this particular system, or to the value for the silicon grain boundary?
2) The 2011 Sheng potential should generally work fine to describe the phononic part of the thermal conductivity. But you can make a check yourself by calculating the phonon bandstructure using the potential and compare to experiment or higher-level calculations.
3) Of course, ATK-Classical does not account for the electronic contribution to thermal transport, and, since the electronic contribution can make up a large part of the total thermal conductivity in metallic systems, the simulated value may be considerably lower than the experimental value of such a system.
You can calculate both phononic and electronic conductance using the NEGF-method, as explained in the tutorial
http://docs.quantumwise.com/tutorials/thermoelectrics_cnt_isotope.html4) In principle it is not suspicious, that your transferred energy is lower than in the silicon example. This value depends on the many factors, e.g. the cross-sectional-area of the system, and of course also the thermal conductance itself. You can try and tune it by changing the parameter "Exchange interval". By decreasing this parameter the transferred energy per time should increase.
