Dear Ku,
As I said, I would not trust the diffusion coefficients from database because of its specific optimization to the typical temperature ranges of Ni-alloys. I don't know whether this is only true for the prefactor or also for the activation energy. Of course you could keep the activation energy and increase the prefactor by manual input, or you could even continue reading from database and modfiy the values by giving a constant factor on the coefficient. The question is how to argue your choice in case of a later presentation or publication of the results.
The tiny compositions printed in the .diff-file come from using "diagonal_dilute": For the flux element, the composition is reduced by a factor of 1E6 in order to suppress off-diagonal terms and to force the diagonal value to be positive (and reasonable), which is not always the case if you just use the diagonal only in high-alloyed systems. Although the compositions are reduced only 1 by 1 (i.e. leaving the corresponding others at the original value), in the output only the reduced compositions appear, which is sometimes confusing (but not an error).
FCC_L12 and FCC_L12#2 in Ni-based alloys stand for the disordered and ordered fcc-phase, and should behave identically if no ordering is possible. Nevertheless, you should use FCC_L12 (or even better FCC_A1) and manually set the major constituents to only Al on all sublattices (Al as matrix component) while creating the .ges5-file. You can see the effects in MICRESS at the end of initialisation, when the "Start Compositions and Limits for quasi-equilibrium" are written to the screen output for all elements in all phases.
You cannot specify the initial composition of fcc by equilibrium from liquid composition, because the domain initially consists only of fcc. You want to set the fcc composition, not the the composition of hypothetically coexisting liquid. However, you could set them for fcc only using "equilibrium 1", which means that phase 1 is specified, while all other phases (if present) are set according to the equilibrium with fcc.
The "artifact" in the liquid and upper solid part, as I said, is due to regular nucleation of liquid. Please bear in mind, that during this initial extremely fast melting there is even not enough time for the liquid phase to homogenize by diffusion.
Regarding your final question, it is true that if separation of composition sets is difficult, the problem should be solved first at the database level (i.e. by defining proper major constituents when creating the .ges5-file). The effect of that is that TQ will (in case of trouble) always start from the corresponding default major composition (given in "Start Compositions and Limits for quasi-equilibrium"), ensuring that iteration ends in the right composition set (this specifically is done by calling the subrouting tqsdmc). Setting initial compositions in MICRESS (i.e. using "input" and thus specifying initial composition even if phases are not yet existing) has only an effect when the phases appear for the first time (so that no composition guess is available). In practice this is only relevant in case of reading initial microstructures from restart file, because here all phases are already present but no phase compositions known.
Bernd