Diffractive combining and control of femtosecond pulse beam arrays

We have developed a scalable, ultrafast laser beam combination scheme, which can combine many beams using two diffractive optics. A feature of this approach is the information contained in the uncombined output beams, which can be used to derive phase error information. We show that a machine-learning algorithm can learn to stabilize beam combination with high efficiency, by finding correlations between uncombined output beam patterns and phase errors.