Learning complexity dynamics in laser-matter interaction through predictive modelling

Ultrafast laser pulses can teach metallic surfaces to adapt. We demonstrate that repeated femtosecond laser irradiation drives a self-organized, cumulative evolution of surface structures that progressively enhance energy absorption, exhibiting learning-like dynamics. Each pulse modifies the morphology, and through feedback-driven reorganization, the surface develops increasing complexity to meet optical constraints. This adaptive behavior emerges from the interplay of thermoconvective instabilities, near-field electromagnetic effects, and dissipative self-organization in the transient molten phase. By combining experimental analysis and neural solvers trained on Maxwell’s equations, we uncover how surfaces evolve toward optimized light-matter interaction. These findings link ultrafast non-equilibrium physics with physical learning processes, opening new pathways for controlled energy deposition and the design of functional, tunable photonic materials.