How to add gradient or hessian matrix for fmincon in yalmip

[NickJay2]

I have a very complicated objective function, which results that the optimization not completed even after 10000 iterations. Actually, I can give the analytical solution of hessian matrix. Will accurate hessian matrix accelerate the optimization? If yes, how can I write the code to add analytical solution of the hessian matrix. Part of my cods is as below.
qx=sdpvar(N,1); qy=sdpvar(N,1); Constraints=[qx(1)==qx_I,qy(1)==qy_I,qx(N)==qx_F,qy(N)==qy_F]; for n=1:N-1 Constraints=[Constraints,((qx(n+1)-qx(n))^2+(qy(n+1)-qy(n))^2)<=(V_max*delta_t)^2]; end obj=0; for n=1:N obj=obj+sum(R_klb(a(n,:),b(n,:),gama0,H,C1,C2,B1,B2,wx,wy,qx(n),qy(n),qx_r(n),qy_r(n)).*... exp(-alpha*R_klb(a(n,:),b(n,:),gama0,H,C1,C2,B1,B2,wx,wy,qx(n),qy(n),qx_r(n),qy_r(n))))/... sum(exp(-alpha*R_klb(a(n,:),b(n,:),gama0,H,C1,C2,B1,B2,wx,wy,qx(n),qy(n),qx_r(n),qy_r(n)))); end Objective=-obj/N; ops = sdpsettings('fmincon.maxfunevals',10000,'fmincon.MaxIter',3000); sol=optimize(Constraints,Objective,ops);

[johanlofberg]

Gradients are automatically computed and supplied to fmincon. Hessians are not supported so fmincon will use quasi-newton if it uses hessian

[NickJay2]

So, is there any method I can take to accelerate it? I may need to repeat the optimization about 1000 times (over 1000*10000 iterations), which is a quite hard task.

[johanlofberg]

Well, you can always try tweaking algorithmic settings in fmincon such as switching algorithm, but most likely you have to come up with a better way to represent the objective. If you can compute gradients and hessians, then it sounds like a pretty simple compact model and you might be better off simply interfacing fmincon directly, avoiding any kind of extra variables etc that are introduced by yalmip