Alright, so I do research in and produce optimization tools professionally. From my biased point of view, someone is better off learning generic optimization t

Alright, so I do research in and produce optimization tools professionally. From my biased point of view, someone is better off learning generic optimization theory and algorithms rather than specialized versions like convex or stochastic optimization. Generally speaking, generic nonlinear optimization methods form the foundation for everything and then there's a series of tricks to specialize the algorithm. Very specifically, there are two books that I think provide a good foundation. First, is Convex Functional Analysis by Kurdila and Zabarankin. Not many people know about this book, but essentially it provides a self-contained background to prove the Generalized Weirstrass Theorem, which details the conditions necessary for the existence and/or uniqueness of an optimization problem. This is important because even convex problems don't necessarily have a minimum. For example, min exp(-x) doesn't have one, but it does have an infimum. The background necessary to understand this book is real analysis and as a quick aside I think Rudin's Principles of Mathematical Analysis is the best for this. Second, is Nocedal and Wright's Numerical Optimization book. It provides a good overview of the powerful algorithms in optimization that we should be using. Now, it's weakness is that it often cheats and uses a stronger constraint qualification than we're afforded in practice. Candidly, I find that the derivative of the constraints will not remain full rank and we will likely violate the LICQ. Further, it covers a number of algorithms that really shouldn't be used in practice, ever. That said, it does cover the good algorithms and it generally has the best presentation out of the other books. Sadly, I don't know of any killer books for numerical linear algebra. And, yes, I've read cover to cover things like Saad's Iterative Methods for Sparse Linear Systems, Trefethen and Bau's Numerical Linear Algebra, and Golub and van Loan's Matrix Computations. They're valuable and well-written, but don't quite cover what I end up having to do to make linear algebra work on my solvers. Anyway, this is all biased and opinion, so take what you will. If someone else has some of their favorite references for optimization or numerical linear algebra, I'd love to hear.