BAND does not calculate the total energy. Instead, it calculates the formation energy with respect to the spherically symmetric spin-restricted atoms. The reason is that this number is easier to determine accurately using numerical integration then the total energy.
The problem is that BAND is (internally) centralizing the atom positions with respect to the geometrical center of the structure. During this centralization an atom can be positioned on one side of the unit cell or the other. (the result for this individual calculation will be the same in both cases, but for a PEDA calculation that is not true) In your case, the centralization of the fragments and of the PEDA calculation gave results which cannot be transformed into one another by a simple translation.
The solution is to pre-centralize the whole system with respect to the geometrical center of the “bigger” fragment. (since this fragment usually gives this problem) E.g. if you study a surface-adsorbate interaction, you centralize the whole structure w.r.t. the geometrical center of the surface. In some cases the adsorbate fragment might be the reason of the problem. If so, we recommend redefining the unit cell so that the adsorbate is closer to the center of the unit cell than to any of the borders of the unit cell.
If this approach does not solve your problem please send us your input and output, so we can look into it.
The dimension of the text-book “density of states per unit cell” (DOS) is [1/energy].
What is plotted in the BAND-GUI is not the DOS, but a histogram of the DOS, which is a dimensionless quantity.
The “unit” of the y-axis is [number of states / unit cell].
Each data-point in the plotted DOS represents the number of states (per unit cell) in an energy interval DeltaE. The histogram bin-width DeltaE depends on the DOS input options “Energies”, “Min”, “Max”: DeltaE = (Max-Min)/Energies.
In case one really wants the “text-book” dos, one can set the Dos%IntegrateDeltaE option to “false”. The result will be a very wild function.
The negative values in the partial DOS are artifacts due to the fact that in a non-orthogonal basis set the definition of a “partial DOS” is somewhat arbitrary.
The partial DOS can be a useful tool for understanding the physics/chemistry of your system, but it’s strictly speaking not a “physical quantity”.