Exploratory FactorAanalysis Heywood Case


I conducted an exploratory factor analysis (n=200, 18 variables) using Maximum Likelihood to estimate the factor loadings. However, my factor analysis failed to converge due to a Heywood solution. Given that my data are non-gaussian and that ML assumes multivariate normal distribution for the indicators I thought that this might be one reason for the Heywood solution. (Is this assumption reasonable?) Thus, I used Iterative Principal Factor (PF) method instead, since this method does not rely on distributional assumptions. The results I receive using PF is much easier to interpret, also the goodness of fit measures improved (Bentler-Bonnet Normed (NFI) and Root mean sq. resid. (RMSR)). However, I am wondering whether using PF instead of ML is in fact an acceptable solution for Heywood cases or if my reasoning is wrong.
Does anybody have an idea? I would very much appreciate any help :)