Convergence of common proximal methods for ℓ₁-regularized least squares

We compare the convergence behavior of ADMM (alternating direction method of multipliers), [F]ISTA ([fast] iterative shrinkage and thresholding algorithm) and CD (coordinate descent) methods on the model ℓ₁-regularized least squares problem (aka LASSO). We use an eigenanalysis of the operators to compare their local convergence rates when close to the solution. We find that, when applicable, CD is often much faster than the other iterations, when close enough to the solution. When far from the solution, the spectral analysis implies that one can often get a sequence of iterates that appears to stagnate, but is actually taking small constant steps toward the solution. We also illustrate how the unaccelerated ISTA algorithm can sometimes be faster compared to FISTA when close enough to the solution.