Lecture 14: Methods (Truncated Newton Method)

So this is what we’ll do. The problem was scaled in such a way that we could pull off a Cholesky factorization. I think the Cholesky factor had something like 30 million nonzeros. So it’ll take some time to do both the Cholesky factorization and also to do the backward and forward substitution. So but direct is possible. All we have to do with this problem is scale it by a factor of ten and direct becomes kind of out of the question, so then at least on a little standard machine. ... See the whole transcript at [[http://see.stanford.edu/materials/lsocoee364b/transcripts/ConvexOptimizationII-Lecture14.pdf|Convex Optimization II - Lecture 14]]