In particle accelerators, particle losses to surrounding components must be minimized. These particles usually have been driven into an undesired "halo" outside the desired distribution before being lost. The processes involved and physical identification of such halo is an intense research topic especially when the nonlinear space charge effect is taken into account. In order to present a method, firmly based on the physics of the transient beam state, for determining when a particle may be defined as being in a halo, the particular example of an initially rms matched beam injected into a resonance of an alternating-gradient (AG) focusing lattice is studied, in which the resonance between the collective mode and single particle motion pushes some particles into a halo defined by extending the halo description to the action-angle frame. In this new frame of description, the halo particles can be physically defined as those with larger actions affected by the outer quasi hyperbolic tori, which contains the halo particles observed in the traditional approaches on projected 2D phase space, as well as some particles still hidden in the core.