search

JuliaFirstOrder / ProximalAlgorithms.jl

Proximal algorithms for nonsmooth optimization in Julia
Other
130 stars 21 forks source link

Question on DavisYin algorithm #83

Closed patwa67 closed 2 years ago

patwa67 commented 2 years ago

I was looking at the Davis and Yin (2017) paper and their description of Algorithm 2. I'm a bit confused how their notation corresponds to the one in the manual. They also use f(x)+g(x)+h(x), but they take the gradient of h(x) whereas in the documentation the gradient is taken of f(x). Can you clarify the relation between the documentation and Algorithm 2?

lostella commented 2 years ago

I went for consistency within the package, so f is always the smooth term whenever possible. Wrt the arxiv version of the Davis-Yin paper then https://arxiv.org/pdf/1504.01032.pdf f takes the place of h in the paper, while g and h are the (interchangeable, as far as modeling is concerned) proximable terms. The published version uses the same notation if I remember correctly (can’t access it now)

patwa67 commented 2 years ago

OK, that's what I suspected. I just wanted to get it confirmed.