There are some very excellent software packages for solving constrained optimization problems (such as IPOPT). My code is not as efficient or as robust or as flexible as others, but it has the virtue of being simple (the solver is only 150 lines long!) and (relatively) easy to understand and use, and still works pretty well. It should also be reasonably fast for optimization problems that aren't too demanding.

My MATLAB code is based on the recent work in interior-point methods, specifically those methods that keep track of both the primal and dual optimization variables (hence primal-dual methods). These methods are special because they are numerically stable under a wide range of conditions, so they should work well for many different types of constrained optimization problems. That being said, you can always find a constrained optimization problem that is difficult enough to break these methods, so be careful! See below for a brief list of references that lay the foundation for this code.

I use a merit function which ensures global convergence (see the references below). What this means for you is that you don't have to use the Newton step (and compute the Hessian of the objective, which might be expensive to compute) to be guaranteed to converge to the optimal solution. For instance, you could use a quasi-Newton approximation to the Hessian instead, which is computed using the gradient information.

If you have any questions, praise, or comments, or would like to report a bug, do not hesitate to contact me. I've tested this software in MATLAB version 7.5.

License

Installation and use

Click here to download a compressed TAR archive containing six MATLAB files. The interior-point solver is ipsolver.m. There are also two files for a demonstration of how to use my MATLAB function to find the solution to a convex quadratically-constrained quadratic program, and there are another three files which comprise a demonstration of logistic regression, again using my implementation of the primal-dual interior-point solver. If you don't know what logistic regression is and how it is used for classification, and how it is connected to constrained, convex programming, I highly recommend reading up on it. See the bottom of this page for some suggested references on this topic.

If you type help ipsolver in the MATLAB console, you will see instructions on how to use the solver for your constrained optimization problem. I also highly recommend running and understanding the two examples I provided, since they should give you a good idea how to use solver in your own MATLAB script.

Some introductory references on interior-point methods and logistic regression

Stephen Boyd and Lieven Vandenberghe (2004). Convex Optimization. Cambridge University Press.

Jorge Nocedal and Stephen J. Wright (2006) Numerical Optimization, 2nd edition. Springer.

Margaret H. Wright (1991). Interior methods for constrained optimization. Acta Numerica, pp. 341--407.

Paul Armand, Jean C. Gilbert, and Sophie Jan-Jégou (2000). A Feasible BFGS Interior Point Algorithm for Solving Convex Minimization Problems. SIAM Journal on Optimization, Vol. 11, No. 1, pp. 199-222.

Trevor Hastie, Robert Tibshirani and Jerome Friedman (2001). The Elements of Statistical Learning. Springer.