Since the vibration mitigation of nonlinear energy sinks (NESs) needs multi-objective optimization, this paper aims to address this issue with efficient multi-objective particle swarm optimization (MOPSO) methods. A non-polynomial NES model, namely a square root NES (SRNES), is targeted for multi-objective optimization and dynamical analysis on an impulsively excited linear oscillator (LO) and compared with the corresponding polynomial model. Different MOPSO methods are developed to balance the mass and dissipation efficiency of SRNESs by utilizing weighted ranking, external archive, and non-dominated sorting. And nonlinear normal modes (NNMs) are extended for the analysis of SRNESs. The optimization results show the efficient targeted energy transfer (TET) and high robustness of the bistable SRNES in different LO initial and damping conditions. Specifically, more than 95% energy absorption can be easily witnessed, even in a small mass ratio (around 0.003), but the amplitude of the SRNES can be very large. The principal manifestation of the optimal TET is 1:1 resonance capture along with considerable low-frequency components of the SRNES. The optimization results of the SRNESs and the linear vibration absorbers show high similarities, while with a large mass ratio, the SRNESs possess broader range of spring stiffness. In the comparison of the two NES models, the differences are found in potential energy surfaces, NNMs, in-well motion, and vibration attenuation performance. Besides, the SRNESs can be divided into two groups in terms of the conservative dynamics, which can be explained by the polynomial model. The frequency component of in-phase NNMs dominates in the conservative and dissipative dynamics of the SRNESs, which lead to the transition of NNMs in the dissipative dynamics. In summary, the efficient MOPSO methods can be designed for different needs of vibration suppression and structure cost. Furthermore, it is important to directly analyze the non-polynomial nonlinear vibration, because transforming non-polynomial models into polynomial ones may lead to unreasonable or imprecise results. Finally, the NES damping structure and the NES performance in small mass conditions should be enhanced.