(INFO
", based on a simulation of anti-podal point pairs repelling each other "
"on the unit sphere surface. This can either distribute more uniformly "
"a given set of gradients, or it can make a new distribution from scratch. "
"A more clever implementation could decrease drag with time, as the "
"solution converges, to get closer to the minimum energy configuration "
"faster. In the mean time, you can run a second pass on the output of "
"the first pass, using lower drag. A second phase of the algorithm "
"tries sign changes in gradient directions in trying to find an optimally "
"balanced set of directions. This uses a randomized search, so if it "
"doesn't seem to be finishing in a reasonable amount of time, try "
"restarting with a different \"-seed\".")