The fundamental equations of classical physics are ordinary or partial differential equations for the observables which describe the state of a system. Newton's equations of classical mechanics, Maxwell's equations of classical electrodynamics, and Einstein's equations of the theory of general relativity describe and predict the state of a system, if the initial and boundary conditions are known. The precision and reliability of these theories were the reason for great scepticism, when the theory of quantum mechanics was developed. Suddenly, it was not possible to calculate the state of a system with arbitrary accuracy. The result of quantum mechanical calculations are probabilities for the happening or non-happening of an event. But also in classical physics there are situations, where it is more adequate to talk about the probability for a certain event or the mean value of an observable, than to calculate the exact trajectories. This is the case when large systems with many degrees of freedom are investigated. Statistical physics uses this paradigm to treat large, complex systems. The well established theory of statistical physics of equilibrium states gives probability distributions for the micro states (characterized by the position and momentum of particles for example), if a certain macro state (defined by energy density, temperature, pressure etc.) is defined.
However, ``real life systems'' are hardly ever in equilibrium. Usually the observed phenomena are time dependent, or, even if they are stationary, the systems are open due to the exchange of energy or another physical quantity.
In order to describe nonequilibrium systems it is necessary to identify those modes which dominate the time evolution of the system. In complex systems there are usually many different subsystems, which can be characterized by their own dynamics. There are fast, slow and almost steady subsystems. If the time scales are far enough apart, the fast subsystem can be treated as a noise. The almost steady subsystem can be treated as static. So, there is only the slow system left, which has to be considered in greater detail.
For the micromagnetic systems which are considered in this thesis, the influence of thermal activation is identified with the fast subsystem and can therefore be treated as noise. The almost steady influence can arise from external magnetic fields. The typical time scale for the dynamics of the magnetization is given by the Larmor precession frequency which is in the order of 
Hz. However, if the switching frequency of external magnetic fields reaches a comparable order of magnitude, it has to be studied in more detail.
This problem is of immediate scientific interest for magnetic recording applications. The increasing data transfer rates and writing speeds require higher frequencies of the applied fields. As the typical frequencies of the recording head currents approach the precession frequency, the process of magnetization switching can be expected to be influenced. The basic theory, algorithms, and programs, which are studied in this thesis, will provide a framework for advanced research and investigations of these phenomena.