Whenever watching sports where the athletes perform somersaults one can observe that the movement of the arms influence the twist rotations. In this article the movement is modeled and simulated with two different versions of the stretched arm motions. Once with a one degree of freedom (DOF) planar motion in the shoulder and once with a 2 DOF motion therein. These two arm strategies were denoted by the 2D and 3D techniques, respectively. The base, i.e. the remaining body, is simulated as a free-floating base with 6 DOFs. The goal of this manuscript is to find an optimal arm strategy for the different degrees of freedom to maximize the number of twists while the body performs a backward somersault. Therefore, an optimal control law was set up that satisfies the dynamics and position, velocity, angle and rotation constraints of the virtual body. Two different cost functions were implemented to show the coherence of complexity between the whole motion and number of twists. Overall in this study it turned out that good 3D techniques yield a much higher number of twists than good 2D techniques, although the complexity of the underlying motor control is not increased. A drawback of the attempt using two degrees of freedom in the shoulder is that the convergence rates in the optimization are lower.