Before you read this, please understand I'm not a specialist in optimization so I may be ignorant of many well-known results in that field.
I am interested in hearing about methods for minimizing the quadratic cost $L(u) = u^{tr}Mu$, where $u \in R^n$ and $M$ is an $n\times n$ positive semidefinite matrix, subject to the constraint $0 < \sum_{i=1}^n|u_i| \leq 1$.
I have a feeling some form of dynamic programming can do this, but I would need to be pointed to some relevant literature.
Edit: I think this is phrased poorly and most people who tried to help are confused. I have tried deleting it and trying again, but it won't allow me to delete due to there being an answer.
While I was not deliberately trying to redefine the problem after someone answered--$u = 0$ is just not reasonable given the application--trying to make myself clear made it turn out that way.
I also don't think the problem I want to communicate is unsolvable; that would again be unreasonable, so I must have screwed up. I am just going to abandon this failure to launch and try again.