VU algorithm
Notes on the VU algorithm of
Mifflin, Sagastizábal (2005) A VU-algorithm for Convex Minimization, Mathematical Programming.
Types
NonSmoothSolvers.VUbundle — TypeParameters:
ϵ: overall precision requiredm: sufficient decrease parameterμmin: minimal prox parameter (μ is inverse of γ). Higher μ means smaller serious steps, but less null steps
NonSmoothSolvers.VUbundleState — TypeParameters:
σ: in (0, 0.5!], lower values enforce higher precision on each prox point approximation,
Methods
NonSmoothSolvers.qNewtonupdate! — FunctionqNewtonupdate!(H, pₖ, pₖ₋₁, sₖ, sₖ₋₁, U, k, curvmin, ν, νlow, μ, kase) -> Tuple{Any, Any, Any}
Update the quasi-Newton H matrix from updates pₖ, pₖ₋₁, sₖ, sₖ₋₁ with BFGS or SR1 scheme.