Reversibility and stability of information processing systems

Classical and quantum models of dynamically reversible computers are considered. Instabilities in the evolution of the classical 'billiard ball computer' are analyzed and shown to result in a one-bit increase of entropy per step of computation. 'Quantum spin computers', on the other hand, are not only microscopically, but also operationally reversible. Readoff of the output of quantum computation is shown not to interfere with this reversibility. Dissipation, while avoidable in principle, can be used in practice along with redundancy to prevent errors.