Landauer's Principle

27 Feb 2017 16:30

A (somewhat disputed) result in statistical mechanics, linking it to computation and information theory, named after Rolf Landauer. (In a shocking violation of Stigler's Law of Eponym, Landauer really does seem to have been the first to articulate it.) The principle is that erasing one bit of information must produce at least \( kT \ln 2 \) joules of heat, where \( T \) is the absolute temperature and \( k \) is Boltzmann's constant. This is held to be because erasing one bit must increase entropy by one bit, i.e., by \( \ln 2 \) nats, and since heat is related to entropy (by \( dS = dQ / T \), erasure must produce heat.

For a long time, I thought that this was both an obviously correct and a very deep argument, but reading some critical papers by Shenker and Norton in 2005 persuaded me that the argument above is much too hasty, and that other arguments were at least doubtful if not invalid. I have remained in a state of doubt, suspended judgment, and curiosity ever since.