I’m familiar with the work by Admal and Tadmor, and I think you need to do some background reading from Hardy’s original paper, or the ones I referred to in my last e-mail. For Hardy’s method of stress, one uses special functions (called localization functions) to perform a weighted spatial averaging of atomic contributions to stress evaluated at a spatial point. This spatial point may coincide with an atom’s position, but doesn’t need to.
The setup of a grid of spatial points, as well as evaluation of those localization function weights is not performed in the normal version of lammps. You would need to compile and use the user-atc library in order to do this, or write you own compute subroutine to perform these evaluations.