Can you post some of your xmu.dat files generated with various cluster sizes? That might give a clue as to what is going on.

If I increase the broadening for the LDOS, the default is 0.1eV for the one exported by Arthemis and 0.2eV is shown in the manual, the last one is still small compared with 4.16 eV. Which is a good criteria to choose it and where I can find the core hole broadening value?

The core-hole broadening in eV is reported in the header of xmu.dat as Gam_ch. Here is the relevant line from one of my files:

# Gam_ch=8.302E+00 H-L exch Vi= 2.000E+00 Vr= 0.000E+00

I will quit the LDOS card to obtain results faster, but pot.bin will be generated anyway (I will leave the CONTROL card option on), if not, is it needed for another calculation? The calculations (without the LDOS) and the spectra will be obtained like before?

You can compute pot.bin once and do subsequent calculations with CONTROL 0 1 1 1 1 1. This will cause FEFF to skip the potentials calculation and use the result (saved in pot.bin) of the previous run. There is no need to recalculate the potentials for each run in a convergence test unless you are converging parameters relevant to the potentials calculation.

Best Wishes,

Micah Prange