academia nuts
so this problem came up at work. i have an approximate solution to minimizing a non-linear function of six variables. but i need to refine it and make it better. seems like someone somewhere has already solved this problem. my google prayers were answered with a bounty. yes indeed. this problem has been studied ad infinitum. most solutions involve the jacobian, the hessian, conjugate gradients, line search, bingo! whew. i'm sure if this problem came up more often i'd have efficient solvers in my toolbox. but it doesn't. and i don't. i tried some of them but quickly ran into convergence failures. which seems really wrong given the well-behaved nature of the evaluation function. eventually, i threw up my hands. screw academia. i wrote a quick and dirty greedy thing. it converges. not optimally. but the result looks like it's as good as can be. and it's fast enough. which ultimately is the criteria for these proof of concept experiments. sheehs.