Coordinate Descent
Main Idea
In Unconstrained Optimization, coordinate descent is similar to applying a Preconditioner to Gradient Descent (see Newton's Method). This preconditioner sets the descent direction to be along just one coordinate / one entry of the vector to be optimized.
Consider
Gradient descent iterates as
Preconditioning with
which is done by