Dual initial solution for primal-dual solvers

[NizarBousselmi]

Hello,

I have a question concerning the initialization of the dual solution in the class "primal_dual". We consider the minimization of f(Lx) + g(x). If I understood well, if the dual initial point is not specified (self.d0 is None) then you set u0 = Lx0 (where x0 is the primal initial point).

My question is why this choice of u0? Isn't it more "natural" to choose u0 as the (sub-)gradient of f at L*x0 ?

Thank you for your very nice toolbox!