学术文献库

Energy consumption of computing has found increasing prominence but the area still suffers from the lack of a consolidated formal theory. In this paper, a theory for the energy consumption of computing is structured as an axiomatic system. The work is pure theoretical, involving theorem proving and mathematical reasoning. It is also interdisciplinary, so that while it targets computing, it involves theoretical physics (thermodynamics and statistical mechanics) and information theory. The theory does not contradict existing theories in theoretical physics and conforms to them as indeed it adopts its axioms from them. Nevertheless, the theory leads to interesting and important conclusions that have not been discussed in previous work. Some of them are: (i) Landauer's principle is shown to be a provable theorem provided that a precondition, named macroscopic determinism, holds. (ii) It is proved that real randomness (not pseudo randomness) can be used in computing in conjunction with or as an alternative to reversibility to achieve more energy saving. (iii) The theory propounds the concept that computers that use real randomness may apparently challenge the second law of thermodynamics. These are computational counterpart to Maxwell's daemon in thermodynamics and hence are named daemon computers. (iv) It is proved that if we do not accept the existence of daemon computers (to conform to the second law of thermodynamics), another type of computers, named clairvoyant computers, must exist that can gain information about other physical systems through real randomness. This theorem probably provides a theoretical explanation for strange observations about real randomness made in the global consciousness project at Princeton University.