It is shown that a unique measure of volume is associated with any statistical ensemble, which directly quantifies the inherent spread or localisation of the ensemble. It is applicable whether the ensemble is classical or quantum, continuous or discrete, and may be derived from a small number of theory-independent geometric postulates. Remarkably, this unique ensemble volume is proportional to the exponential of the ensemble entropy, and hence provides a novel geometric characterisation of the latter quantity. Applications include unified volumebased derivations of the Holevo and Shannon bounds in quantum and classical information theory, a precise geometric interpretation of thermodynamic entropy for equilibrium ensembles, a geometric derivation of semi-classical uncertainty relations, a new means for defining classical and quantum localization for arbitrary evolution processes, a geometric interpretation of relative entropy, and a new proposed definition for the spot-size of an optical beam. Advantages of ensemble volume over other measures of localization (root-mean-square deviation, Renyi entropies, and inverse participation ratio) are discussed.PACS Numbers: 03.65. Bz, 05.45+b, 42.60.Jf
I INTRODUCTIONThis paper has two main goals. The first is to demonstrate that for any ensemble, whether classical, quantum, discrete or continuous, there is essentially only one measure of the "volume" occupied by the ensemble which is compatible with basic geometric notions. This ensemble volume is thus a preferred and universal choice for characterising what is variously referred to as the spread, dispersion, uncertainty, or localisation of an ensemble. Remarkably, the derived "ensemble volume" turns out to be proportional to the exponential of the entropy of the ensemble. A by-product of the first goal is thus a new universal characterisation of ensemble entropy, based on geometric notions. Indeed, a number of properties of ensemble entropy turn out to have simple geometric interpretations. The universal nature of the characterisation is of particular interest: the only previous contextindependent interpretation of ensemble entropy to date (and hence applicable in particular to ensembles described by continuous probability distributions) appears to be as a somewhat vague measure of uncertainty or randomness.
2The second goal is to apply "ensemble volume" to a wide range of contexts in which ensembles appear. The applications demonstrate not only the advantages of ensemble volume over other measures of spread, but also to some extent why it is that ensemble entropy makes a natural appearance in contexts as diverse as statistical mechanics, information theory, chaos, and quantum uncertainty relations. Some results have been briefly reported elsewhere [1]. Here important details and extensions are given, as well as a number of new results.The work reported here was originally motivated by several connections between volume and information. Shannon proved an upper bound on information transfer, via classical signals s...