Sorry for the delay with answer.
The convergence that you mentioned is _not_ that of the iterative solver when computing the internal field on the
dipoles (which is characterized by relative residual norm. Instead it is the convergence of the Romberg integration,
which is used either for orientation averaging or for integration of the scattered field over the whole solid angle
(e.g. to compute asymmetry parameter) - see section "Romberg integration" (11.5) of the manual. This integration routine
provides an estimate of the final accuracy, usually this estimate is larger than the real error.
The output that you got means that ADDA can not guarantee that the integration was performed accurate enough. You can
either ignore this message (if you have other means to test the accuracy) or try increasing values of Jmax in file
avg_params.dat or alldir_params.dat (see corresponding sections of the manual - B.2 and B.3).
Maxim.