An Agnostic Machine Learning Model of Photosynthetic Habitability

Traditional estimates of the Photosynthetic Habitable Zone (PHZ) are based on empirical data from Earth-based phytoplankton, which may impose an 'Earth-centric' bias on our search for extraterrestrial life. This paper asks: what are the fundamental thermodynamic and chemical limits of photosynthesis if we remove these biological assumptions and instead model photosynthesis as an optimized process driven by natural selection?

The authors developed an agnostic machine learning model that treats photosynthesis as a generalized photochemical reaction. Instead of using specific organisms, they modeled the light-harvesting antenna as a system of pigments optimized by a genetic algorithm. This system is constrained only by the laws of thermodynamics and redox chemistry. The model simulates how these 'learning antennae' would adapt to the spectral irradiance of planets orbiting various main-sequence stars, allowing for both oxygenic and anoxygenic photosynthetic pathways.

The simulations reveal that photosynthetic organisms can compensate for lower stellar flux by evolving larger, more complex light-harvesting structures. Because of this adaptive capacity, the viability of photosynthesis declines only linearly with orbital distance, rather than following the quadratic drop-off of stellar flux. Consequently, the PHZ is much broader than previously thought. While Earth-like oxygenic photosynthesis is limited for cool M-dwarf stars, the model shows that anoxygenic or alternative NIR-driven oxygenic photosynthesis could be robustly viable across the entire habitable zone of these stars. This suggests that M-dwarf planets could host life that produces unique reflectance biosignatures in the near-infrared (NIR) band.