We present an algebra with a fixpoint operator which is suitable for modeling computations with distributed collections found in big data frameworks. We show that under reasonable conditions this fixpoint can be evaluated by parallel loops with one final merge rather than by a global loop requiring network overhead after each iteration. We also show when and how filters can be pushed through recursive terms, proposing optimisation rules. This makes it possible to develop algebraic optimizations in the presence of recursion, in a way which is especially suited for query optimizers and compilers targeting big data frameworks.