Gambling with Heat: A new engine that outperforms Carnot

• The study, published in Physical Review Letters, introduces the “Gambling Carnot Engine”, a feedback-controlled heat engine that can surpass Carnot’s efficiency and achieve full conversion of heat into work under realistic conditions.
• The research is the result of a collaboration between the ICTP (Italy), École Normale Supérieure (France), the University of Potsdam (Germany), and IFISC (UIB-CSIC, Spain).

Researchers have proposed a novel type of microscopic heat engine that challenges one of the oldest principles of thermodynamics. By merging insights from gambling strategies with advanced feedback control, the so-called Gambling Carnot Engine (GCE) can extract work from heat at efficiencies beyond the Carnot bound, the traditional theoretical maximum for any classical heat engine.

The study, published in Physical Review Letters, introduces an experimentally realizable feedback protocol applied to a Brownian Carnot engine, a microscopic machine in which a colloidal particle suspended in a fluid is confined and manipulated by external forces. Unlike traditional engines, the GCE combines thermal resources with information about the system’s random fluctuations to decide the optimal moment for intervention, in a way similar to gamblers placing bets under favorable odds. This smart timing allows the engine to enhance both power and efficiency, approaching perfect heat-to-work conversion in the ideal limit of infinitely slow cycles.

“Our work shows that information can be used as a physical resource to bypass traditional thermodynamic limits”, says IFISC (CSIC-UIB) researcher Gonzalo Manzano. “By applying a feedback rule inspired by gambling, we can design engines that not only outperform Carnot’s efficiency at maximum power, but also, in principle, approach 100 percent efficiency in ideal conditions”.

At the heart of the mechanism lies a “gambling demon”, an external controller that continuously monitors the random motion of a Brownian particle. When a predetermined condition is met, namely the particle crossing a critical position, the engine performs a sudden and cost-free adjustment. This intervention reduces energy waste and boosts overall performance. The strategy makes use of mathematical tools from martingale theory, which describe the statistics of random events, much like analyzing the odds in games of chance.

The authors validated their theoretical predictions with numerical simulations, which confirmed that the GCE consistently outperforms its classical counterpart under realistic laboratory conditions. Notably, the engine’s efficiency not only surpasses the Carnot limit, but also exceeds the Novikov–Curzon–Ahlborn bound, a benchmark long considered the ceiling for efficiency at maximum power in conventional engines.

Importantly, the researchers emphasize that their findings do not violate the second law of thermodynamics. Instead, they extend its reach: when information about a system is taken into account as a genuine resource, new possibilities arise for extracting useful energy. “The Gambling Carnot Engine is more than a theoretical curiosity”, adds Manzano. “Its principles could inspire a new generation of microscopic machines, where information-driven control enables levels of performance previously thought impossible”.

Beyond statistical physics, the research underscores the growing role of information as a thermodynamic currency. The results open avenues to experimental demonstrations of gambling-inspired feedback in colloidal systems, and may influence the design of microscopic devices for energy harvesting, sensing, and computation at the edge of physical limits.

Looking ahead, the team plans to investigate the trade-offs between the energetic benefits of such engines and the costs of acquiring information at high speed. Understanding this balance will be crucial for developing practical nanotechnologies powered by fluctuations.

T. Tohme, V. Bedoya, C. di Bello, L. Bresque, G. Manzano, and É. Roldán, Gambling Carnot Engine, Physical Review Letters 135, 067101 (2025). DOI: 10.1103/w8cx-xx1z