We provide a mathematically rigorous framework to study the use of Deep Mutual Learning in the context of conditional diffusion models for long tailed generation.