Biological systems are among the most challenging subjects for theoretical physicists, as well as experimentalists or simulationists. Physical principles should have been both constraints and guide-lines for the evolution of living systems over billions of years. One generally aims at clarifying what physical principles, possibly new ones, are behind the phenomena of biological interest and at understanding how they work within the entire biological world. In the present talk we describe an example of such an effort. Since the discovery of `superprecipitation' by Szent-Gyorgyi's group in 1940's, it has been a long puzzle how an assemblage of actin filaments with random orientation can contract in the presence of the two-headed myosin molecules undergoing actin-activated ATP-hydrolysis reaction. It is widely accepted that during the contraction the two-headed myosin mediates the relative sliding of two actin filaments whose polarity directions are not parallel but rather anti-parallel. But this fact solely does not account for the shortening. We propose a dynamical model which, upon numerical simulation, exhibits the shortening of an bundle of the actin filaments which are initially dirstributed randomly both in space along a line and in polarity direction. In the course of shortening several clusters of actins appears along the bundle. The model also shows the sorting of the actin filaments according to their polarity in the late stage. These findings are in accordance with the recent experiment by Takiguchi.