1: \begin{abstract}

2:     We consider the problem of large-scale Fisher market equilibrium computation through scalable first-order optimization methods. 

3:     It is well-known that market equilibria can be captured using structured convex programs such as the Eisenberg-Gale and Shmyrev convex programs.

4:     Highly performant deterministic full-gradient first-order methods have been developed for these programs. 

5:     In this paper, we develop new block-coordinate first-order methods for computing Fisher market equilibria, and show that these methods have interpretations as t\^atonnement-style or proportional response-style dynamics where either buyers or items show up one at a time.

6:     We reformulate these convex programs and solve them using proximal block coordinate descent methods, a class of methods that update only a small number of coordinates of the decision variable in each iteration.

7:     Leveraging recent advances in the convergence analysis of these methods and structures of the equilibrium-capturing convex programs, we establish fast convergence rates of these methods.

8: \end{abstract}

9: