Information-Powered Heat Engine

Information-powered heat engine is a thermodynamic device that extracts usable work not from a temperature gradient, as in a conventional heat engine, but from information about the microscopic state of a system. The paradigm was established by Leo Szilard's 1929 Szilard engine and generalized by the modern framework of thermodynamics of computation: information, when coupled to a feedback mechanism, is a thermodynamic resource on equal footing with heat.

The defining characteristic of an information-powered engine is that its working substance is not a gas or fluid but a probability distribution. The engine operates by measuring the state of a system — typically a single particle in a fluctuating environment — and using that information to modify the system's constraints in a way that extracts work. The measurement provides the 'fuel'; the feedback provides the 'engine cycle.'

The thermodynamic efficiency of such engines is bounded not by the Carnot limit but by the Landauer limit: the work extractable from one bit of information is at most k_B T ln 2, and the cost of erasing the measurement record must be paid to complete the cycle. Information-powered engines are therefore not perpetual motion machines. They obey the second law precisely because the information they consume must eventually be destroyed, and that destruction has a thermodynamic cost.

Experimental realizations include Brownian ratchets that use optical feedback to sort particles, and quantum versions that exploit measurement back-action to drive a piston. These devices blur the line between thermodynamics and computation: the engine is a computer that uses information as fuel, and the computer is an engine that uses logic to pump energy.

The information-powered heat engine is often dismissed as a curiosity of statistical mechanics, a niche result with no practical application. This misses the point entirely. Every neural network that uses prediction to reduce energy consumption, every control system that uses sensor data to optimize a process, every market that uses information to reallocate resources — these are all information-powered engines. The formalism of Szilard and Landauer is not a footnote to thermodynamics; it is the thermodynamics of the information age.