Optimization, when to stop

[Aren Siekmeier]

Sometimes we have to forgive the theoretically minded for exploring the realm of possibilities.

In your situation, your argument makes sense. But what I think Ether was after is that OPR can be computed much more quickly, and perhaps this makes it possible to do that computation much more often for a higher reward. Such as in projections, or monte carlo simulations, or in navigating some some sort of decision tree (ok so I'm making some things up here). But in fact, linear solvers constitute a large field in their own right, enjoying applications in embedded electronics, supercomputing, computational science research, computer vision, artificial intelligence, engineering simulations, and more and more and more. The point is, you can certainly stop optimizing once your particular application no longer benefits from further optimization, in fact your boss may compel you to. But other applications may become feasible if you can massively reduce the cost of the process.

After all, how many quotes are there about never needing 100 KB, oh wait nevermind 100 MB, oh wait nevermind 100 GB of storage space? Similarly for internet speeds, global travel times, the cost of goods and services, etc. When you need a certain result, do what gets the job done. But a little research into how to do it better can put much bigger results on the table.