Optimization help: porting fmincon() to Julia

[longemen3000]

Hi, maybe this can help:

Optim: gradient and Hessian of constraints

Hello, I am using Optim.jl to solve a constrained optimization problem. The gradient is not specified, so finite differences are the default. This works nicely for the objective, but not for the constraints. Thank you! Here ist my mwe: using Optim, Test function fun(x) # objective (1.0 - x[1])^2 + 100.0 * (x[2] - x[1]^2)^2 + (x[3]-x[1])^2 end function con_c!(c, x) c[1]= x[1]^2 + x[2]^2 # 1st constraint c[2]= x[2]* sin(x[1])-x[1] # 2nd constraint c end function c…

Basically:

• cons is a function that,given a vector of variables,gives a vector of constraints (inplace)
• con_Jacobian is the Jacobian of cons
• con_hessian is a little different,it basically is an inplace adding of the Lagrange multipliers

Also, there is AD implemented for constraints, you can try using:
TwiceDifferentiableConstraints(c!, lx,ux,lc,uc,autodiff=:forward) (the documentation seems to lack behind, as the support of AD in constraints was in NLSolversBase.jl, not in Optim)