However, all is not lost, because you might be able to simplify your problem into something actually soluble.
First: are you absolutely sure that you want to optimize over the boundary of the unit ball (this is what the constraint x'x=1 seems to impose)? Isn't it enough to only assure that x is constrained into a sane region, like e.g., the n-dimensional unit box? If it is, you get a quadratic program:
See more on quadratic programs here: Quadratic programming.max_x (x' D x) s.t. -1 <= x_i <= 1 \forall i=1..n
Second: all depends on whether or not your D is negative definite or not (all eigenvalues are negative). If not, you are out of luck: the problem is (at least) NP hard. If, on the other hand, D is nicely negative definite, then the objective function is concave and the uniqueness of the optimal solution is guaranteed. In this case, you could use a commercial quadratic solver, like CPLEX, or, after suitable linearization, a linear program solver. too. E.g., this one Math::GLPK has a nice Perl interface (believe me it's nice: I wrote it!...:-).
In reply to Re: Help with Matrix math!
by rg0now
in thread Help with Matrix math!
by madslynge
| For: | Use: | ||
| & | & | ||
| < | < | ||
| > | > | ||
| [ | [ | ||
| ] | ] |