Using Operator Splitting Methods

The differential equations we get from gradient flow for composite optimization methods split in the sense of splitting methods from ODEs. Can we use Lie-Trotter-Godunov type algorithms to accelerate them?

1. https://sciml.github.io/OrdinaryDiffEqOperatorSplitting.jl/dev/topics/time-integration/
2. https://hal.science/hal-00557593/document

This could be particularly effective with CliqueTrees integration, since those methods require a linear ordering on the suboperators, which could be generated with the matrix reordering techniques.