Can't tell much other than the BFGS approximated Hessian may be resulting in some incorrect direction. It may also be due to the interior point method options (initial barrier, scaling) but its hard to say. Using the exact Hessian, if available, is always better as (i) the sparsity of the Hessian is maintained (BFGS results in a dense Hessian due to the low rank updates), and (ii) BFGS computes an 'approximated' Hessian.
Good luck, Shri You might also want to compare your computed Hessian with a finite-differencing one to ensure it is correct. From: amel zerigui <[email protected]<mailto:[email protected]>> Reply-To: MATPOWER discussion forum <[email protected]<mailto:[email protected]>> Date: Mon, 10 Feb 2014 18:33:39 +0000 To: MATPOWER discussion forum <[email protected]<mailto:[email protected]>> Subject: RE: Hessian problem Thanks Shri, I am using BFGS via Fmincon function. The problem is solved correctly by using Active set and also by using Fmincon finite-difference Hessian estimation. But using BFGS hessian estimation does not solve the problem. Is there any suggestion. (after I will try with exact Hessian). Thanks ________________________________ From: [email protected]<mailto:[email protected]> To: [email protected]<mailto:[email protected]> Subject: Re: Hessian problem Date: Mon, 10 Feb 2014 18:17:58 +0000 Are you using BFGS approximated Hessian via the fmincon function or computing it yourself? What happens if you use your computed Hessian with the interior point scheme in fmincon? Shri From: amel zerigui <[email protected]<mailto:[email protected]>> Reply-To: MATPOWER discussion forum <[email protected]<mailto:[email protected]>> Date: Mon, 10 Feb 2014 14:38:03 +0000 To: MATPOWER discussion forum <[email protected]<mailto:[email protected]>> Subject: RE: Hessian problem But he problem can be solved with Active set option correctly, by using the same stating point. Thanks ________________________________ From: [email protected]<mailto:[email protected]> Subject: Re: Hessian problem Date: Mon, 10 Feb 2014 09:31:56 -0500 To: [email protected]<mailto:[email protected]> It is impossible for me to know. It may in fact not be a problem with either, but rather that the problem you specified is infeasible. But, as I said, I have no way of knowing. -- Ray Zimmerman Senior Research Associate B30 Warren Hall, Cornell University, Ithaca, NY 14853 phone: (607) 255-9645 On Feb 10, 2014, at 8:44 AM, amel zerigui <[email protected]<mailto:[email protected]>> wrote: Hi, I changed the equality constraints of matpower by adding new constraints, in same way I changed also the Jacobian matrix, and when I checked the computed first derivative, I found this result: Nonlinear equality constraint derivatives: Maximum relative discrepancy between derivatives = 4.44985e-009 This mean the modification is correct, then I estimated the hessian by using BFGS but the optimal point found is: No feasible solution found. I would like to know if the problem is Matlab problem or Matpower. Thanks for help.
