"I'm sure that there is a lot of good material in that book, especially on core computer science topics."
If you look carefully, then you will find that as my first sentence.
I'm "aware" is about me, which is not appropriate, instead of about the book, which is appropriate, but, as explained early in the book, nearly no course covers the whole book.
But I was kind to the book: It made an even worse mess out of probability, and that material was farther from the back cover. Also see another post on this thread saying that generally in the book the proofs are poorly written.
"For those skimming the book, or perhaps more commonly, keeping it as a reference, it's a great compendium of algorithms references, much like a slightly more rigorous, paper version of Wikipedia." I can believe that, but it doesn't conflict with what I wrote. Indeed, I found and downloaded a PDF of the book for just the purpose you mentioned and that I mentioned in my first sentence: The book likely has some core computer science stuff, that is, not just poorly done, hijacked applied math that apparently I know much better than the authors, that I may be able to use as I finish the software for my server farm.
Gads, it looks like they want to use the approximation approach of starting with a minimum spanning tree to solve the traveling salesman problem in the plane. The idea also works in any Euclidean R^n. It's a CUTE idea. I first saw it in a paper of Karp who easily has some nice proofs of some astounding properties. Sadly in the book I saw some references for the idea but not from Karp, and I didn't see mention of some of the really nice properties. There's a chance of a biggie point there: For all the teeth grinding over NP-complete, that problem is NP-complete but surprisingly easy to solve with surprisingly good accuracy; so, the suggestion is that there is a crack in the armor of the NP-complete difficulty. That suggestion is actually more important than the problem itself which in practice people are not struggling with. Yes, the traveling salesman problem remains important, say, for production scheduling, but there the problem is not on a Euclidean R^n.
Some computer science is nice and useful. Hopefully there has been some such since I put down Knuth's TACP! So, maybe in some issues of handling deadlocks, transactions, race conditions, reliability via parallelism, and more, there will be something useful in the book it would take me more than 90 seconds to reinvent!
Still I'm correct: They should clean up their act in their math. We should get better from MIT.