The Izumi--Szilard Engine: Separating Information Erasure from Mechanical Work Extraction

Abstract

The conventional Szilard engine combines one-molecule work extraction with measurement, memory, feedback, and memory reset. The Izumi--Szilard engine is introduced here to separate these informational operations from the mechanical work-extraction problem. A symmetric coupling allows either molecular branch to drive the piston while raising the same load, so that no measurement of the occupied side and no branch-dependent feedback protocol are required. The piston can nevertheless become correlated with the molecular branch. The central point is that this correlation is carried by the same physical piston--load coordinate whose energy, entropy, and return dynamics are already included in the thermodynamic balance. An additional universal Landauer reset cost therefore cannot be assigned independently to the disappearance of the branch label. For the defined architecture, \[ W_{\mathrm{erase}}^{\mathrm{separate}}=0 . \] This result does not establish positive net work. Partition operation, finite-mass piston fluctuations, reverse motion, coupling, timing, synchronization, dissipation, and complete statistical restoration of the apparatus must all be included explicitly. We formulate the corresponding signed work balance and propose mechanical and calorimetric tests of complete cycle closure. If the full autonomous dynamics instead yields nonpositive net work, the physical compensation mechanism should be identifiable. If repeated operation restores every internal and auxiliary degree of freedom while producing persistent positive free-energy gain in an external repository from a single thermal reservoir, the result would contradict the Kelvin--Planck statement under the stated physical assumptions and would compel a re-examination of its applicability to this class of system.***************** Version 2: The manuscript has been substantially revised for clarity and journal submission. The theoretical argument and mechanical accounting have been refined, figures have been revised, and statements on AI-assisted tools and publication declarations have been added. The central scope and conclusions of the manuscript remain unchanged.