Abstract We propose a new way of looking at the quantum Maxwell’s demon problem in terms of conditional action. A “conditional action” on a system is a unitary time evolution, selected according to the result of a previous measurement, which can reduce the entropy of the system. However, any conditional action can be realized by an (unconditional) unitary time evolution of a larger system and a subsequent Lüders measurement, whereby the entropy of the entire system is either increased or remains constant. We give some examples that illustrate and confirm our proposal, including the erasure of N qubits and the Szilard engine, thus relating the present approach to the Szilard principle and the Landauer principle that have been discussed as possible solutions of the Maxwell’s demon problem.
We begin with a study of operations and the effects they measure. We define the probability that an effect
a occurs when the system is in a state
ρ by
Pρ(a)=Tr(ρa). If
Pρ(a)=0 and
I is an operation that measures
a, we define the conditional probability of an effect
b given
a relative to
I by
Pρ(b∣a)=Tr[I(ρ)b]/Pρ(a). We characterize when Bayes' quantum second rule
Pρ(b∣a)=Pρ(a)Pρ(b)Pρ(a∣b) holds. We then consider Lüders and Holevo operations. We next discuss instruments and the observables they measure. If
A and
B are observables and an instrument
I measures
A, we define the observable
B conditioned on
A relative to
I and denote it by
(B∣A). Using these concepts, we introduce Bayes' quantum first rule. We observe that this is the same as the classical Bayes' first rule, except it depends on the instrument used to measure
A. We then extend this to Bayes' quantum first rule for expectations. We show that two observables
B and
C are jointly commuting if and only if there exists an atomic observable
A such that
B=(B∣A) and
C=(C∣A). We next obtain a general uncertainty principle for conditioned observables. Finally, we discuss observable conditioned quantum entropies. The theory is illustrated with many examples. Quanta 2024; 13: 1–10.
Abstract In order to reconcile the entropy reduction of a system through external interventions that are linked to a measurement with the second law of thermodynamics, there are two main proposals: (i) The entropy reduction is compensated by the entropy increase as a result of the measurement on the system (“Szilard principle"). (ii) The entropy reduction is compensated by the entropy increase as a result of the erasure of the measurement results (“Landauer/Bennett principle"). It seems that the LB principle is widely accepted in the scientific debate. In contrast, in this paper we argue for a modified S principle and criticize the LB principle with regard to various points. Our approach is based on the concept of “conditional action", which is developed in detail. To illustrate our theses, we consider the entropy balance of a variant of the well-known Szilard engine, understood as a classical mechanical system.
In order to develop a statistical theory of quantum measurements including continuous observables, a concept of a posteriori states is introduced, which generalizes the notion of regular conditional probability distributions in classical probability theory. Its statistical interpretation in measuring processes is discussed and its existence is proved. As an application, we also give a complete proof of the Davies-Lewis conjecture that there are no (weakly) repeatable instruments for non-discrete observables in the standard formulation of quantum mechanics, using the notion of a posteriori states.
It is widely believed that measurement is accompanied by irreversible entropy increase. This conventional wisdom is based in part on Szilard’s 1929 study of entropy decrease in a thermodynamic system by intelligent intervention (i.e., a Maxwell’s demon) and Brillouin’s association of entropy with information. Bennett subsequently argued that information acquisition is not necessarily irreversible, but information erasure must be dissipative (Landauer’s principle). Inspired by the ensuing debate, we revisit the membrane model introduced by Szilard and find that it can illustrate and clarify (1) reversible measurement, (2) information storage, (3) decoupling of the memory from the system being measured, and (4) entropy increase associated with memory erasure and resetting.