Neal Anderson
University of Massachusetts Amherst

Landauer's Limit and the Physicality of Information

A resurgence of interest in Landauer's Principle has been stimulated by recent experimental probes, theoretical progress in nanoscale thermodynamics, and the continued push for computation at extreme energy efficiencies. However, the nature of the underlying connection between information loss and dissipation in computation - and even its necessity - remains stubbornly controversial. One persistent source of this controversy, I argue, is lack of a clear and consistent definition of information that is thoroughly physical on the one hand and compatible with notions of information used in digital computation on the other. Abstract mathematical measures - absent an explicit physical grounding - fail on the first count, while widely used physical self-entropy and observer-independent mutual information measures fall short on the second.

In this talk I advocate for an alternative conception of physical information that closes this gap, introduce an associated quantitative information measure, and show how this measure enables generalization of Landauer's Principle and clarification of the dissipative origins of irreversible information loss in computational contexts. I focus specifically on the controversial distinction between irreversible (unconditional) erasure of unknown data and reversible (conditional) erasure of known data, and show how our conception of information clarifies this distinction and enables straightforward physical accounting of differences in the respective erasure costs. Finally, I discuss advantages for determination of irreversibility induced energy efficiency limits in computing scenarios far more complex than the idealized one-bit memories routinely employed in investigations of Landauer's Principle.