In this paper we extend the notion of "filtration-consistent nonlinear expectation" (or "-consistent nonlinear expectation") to the case when it is allowed to be dominated by a g-expectation that may have a quadratic growth. We show that for such a nonlinear expectation many fundamental properties of a martingale can still make sense, including the Doob-Meyer type decomposition theorem and the optional sampling theorem. More importantly, we show that any quadratic -consistent nonlinear expectation with a certain domination property must be a quadratic g-expectation as was studied in [J. Ma, S. Yao, Quadratic g-evaluations and g-martingales, 2007, preprint]. The main contribution of this paper is the finding of a domination condition to replace the one used in all the previous works (e.g., [F. Coquet, Y. Hu, J. Mémin, S. Peng, Filtration-consistent nonlinear expectations and related g-expectations, Probab. Theory Related Fields 123 (1) (2002) 1-27; S. Peng, Nonlinear expectations, nonlinear evaluations and risk measures, in: Stochastic Methods in Finance, in: Lecture Notes in Math., vol. 1856, Springer, Berlin, 2004, pp. 165-253]), which is no longer valid in the quadratic case. We also show that the representation generator must be deterministic, continuous, and actually must be of the simple form g(z)=[mu](1+z)z, for some constant [mu]>0.
