Customer Reviews

Numerical Recipes 3rd Edition: The Art of Scientific Computing

5 star:		(8)
4 star:		(0)
3 star:		(2)
2 star:		(2)
1 star:		(4)

 

 

Bottom line up front: Every computational scientist should own a copy of Numerical Recipes but, if you already own a previous version, then don't bother upgrading.

I already owned a copy of "Numerical Recipes in C, 2nd Edition" (from 1992), so I was absolutely thrilled when I saw that the book had been updated in over 15 years. This is why I was so underwhelmed with the 3rd edition. As a previous reviewer noted, the vast majority of the book is largely unchanged.

As in previous editions, the authors do a great job of providing codes that cover the spectrum of topics encountered by researchers. As in previous editions, the authors still take the "give a man a fish" instead of the "teach a man to fish" method. This might seem like a negative but, in my opinion, this is why every scientist should own a copy of Numerical Recipes. Often, topics pop up that need immediate solving and one can often find a code for the topic in Numerical Recipes. As in previous editions, Numerical Recipes is really just an annotated code repository, with very stringent/restrictive licensing rules by the way!

However, as the authors note in the introduction, they made a conscious decision to fill pages with verbatim codes, not building insight into various topics. In my experience, the codes given in Numerical Recipes get the job done, but these tend to be simple and less efficient than other well-known algorithms.

As in previous editions, Numerical Recipes is a terrible pedagogical text. If you're interesting in understanding a particular topic, then get a special-purpose book.

Fifteen years after its previous edition, this peerless book on scientific computing has been upgraded with some very welcome changes. Not only have some advances in scientific computing been incorporated, the explanations are even clearer and more detailed than before. More importantly, the code has been reworked so that it is better than it was in the previous edition. I don't agree with the other reviewer that "it is getting worse". However, it still does seem like C++ code that was written by a Fortran programmer who just doesn't want to let go of the past, although I'd have to say that the code has broken away from the Fortran-like structure of previous editions to some degree. If you do scientific computing at all, this new edition is a must have. Below I detail what is different in this new third edition versus the previous 1992 edition. There are a very few sections that were deleted. I don't mention them. Instead I list any sections or chapters that have been added.

1. Preliminaries
Completely reorganized to reflect the book.

2.Solution of Linear Algebraic Equations
THE SAME

3. Interpolation and Extrapolation
3.7 Interpolation on a Scattered Data in Multidimensions
3.8 Laplace Interpolation

4. Integration of Functions
4.5 Quadrature by Variable Transformation
4.8 Adaptive Quadrature

5. Evaluation of Functions
THE SAME

6. Special Functions
6.10 Generalized Fermi-Dirac Integrals
6.11 Inverse of the Function xlog(x)
6.14 Statistical Functions

7. Random Numbers
7.2 Completely Hashing a Large Array
7.3 Deviates from Other Distributions
7.4 Multivariate Normal Deviates
7.5 Linear Feedback Shift Registers
7.6 Hash Tables and Hash Memories

8. Sorting
THE SAME

9. Root Finding and Nonlinear Sets of Equations
THE SAME

10. Minimization or Maximization of Functions
10.1 Initially Bracketing a Minimum
10.6 Line Methods in Multidimensions
10.11 Linear Programming: Interior-Point Methods
10.13 Dynamic Programming

11. Eigensystems
11.2 Real Symmetric Matrices
11.6 Real Nonsymmetric Matrices

12. Fast Fourier Transform
THE SAME

13. Fourier and Spectral Applications
THE SAME

14. Statistical Description of Data
14.7 Information-Theoretic Properties of Distributions

15. Modeling of Data
15.8 Markov Chain Monte Carlo
15.9 Gaussian Process Regression

16. Classification and Inference (NEW CHAPTER)

17. Integration of Ordinary Differential Equations
17.7 Stochastic Simulation of Chemical Reaction Networks

18. Two-Point Boundary Value Problems
THE SAME

19. Integral Equations and Inverse Theory
THE SAME

20. Partial Differential Equations
20.7 Spectral Methods

21. Computational Geometry (NEW CHAPTER)

22. Less-Numerical Algorithms
22.1 Plotting Simple Graphs

I'm a fan of this book since I've been using this book for a very long time. I pre-ordered the new version and got it a week ago. I think the contents are improved after I had a look at it. I'm pretty happy about that. However, the quality of the source codes, well, I have to say it is getting worse. As you may notice, authors of N.R. put a stringent license on usage of their codes, which is fine since these codes are their intellectual properties. But since they are selling their codes, they are supposed to hire some professional programmers to design a beatiful architecture, a nice data structure, and an easy-to-use interface, and implement all the algorithms with efficiency. As I can tell, C++ is abused in the 3rd version in a very bad way. I've been developing scientific computing software using C/C++ over 10 years, and I have to say the authors of the codes organized their work in a weird way. In the previous version of their codes in C++, global variables are still defined and used at so many places. Any professional programmer knows how bad such a programming style is. In this version, instead of wrapping their routines in classes, they simply use "struct" to hold global variables, does this delight you? This is just an example which upsets me. The only good thing is that they finally learned to use template...

As other reviewers have mentioned, this is basically an annotated code repository of solutions to specific algorithmic problems, and the algorithms are good. However, if you want to *use* these solutions in your products, forget it. You'll need to pay (thousands of dollars per year, potentially) for the privilege of an institutional license, and even then you can't incorporate any of the algorithms into a commercial software product. The code is therefore useless. Worse than useless, actually, because if a company owns the book and then uses an algorithm contained in it - even if derived from a different source - it runs the risk of getting sued for licensing violations because they've seen the book. No thanks.

[...]

If the authors went with some kind of traditional open-source license instead, that would be terrific. Right now, it looks like financial greed has gotten in the way of the dissemination of good ideas.

People, stop buying and using this book! The copyright is prohibitive. It is not that they are greedy and want to make money from it. They won't sell you the rights to use any of these algorithms in commercial products, period. They are just stupid.

And don't give your code to a friend or coworker. You just violated the copyright.

Several coworkers have given me simulations with NR code buried in it. I can't use them. It is ILLEGAL!

Stop! Stop! Stop!

Use the GNU Scientific Library. It is free. And legal! And there is a free book on it. Use anything but NR.

The book is ok but the license is terrible. I never seen a book that does not allow you to use the code. I would have known this before, I would had not purchased the book. I might as well use the GNU Scientific Library.

I really do not understand why this is still popular. The codes are buggy and slow. I do not think anybody can use it for real high performance computing. The book is OK for educational purposes. My suggestion is go for BLAS, LAPACK (MKL is the best in the market, but not free. ACML is a free one from AMD.) or GSL.

This is a classic text in numerical computation and should be in the collection of any scientist or engineer working on numerical codes. While an academic researcher, I have used many, many code snippets from this book in my own work.

There are two main downsides of this book: one major, and one minor. The major one is that, for many bread-and-butter numerical computation tasks (such as ODE integration or linear algebra routines) it probably makes more sense to use one of many free libraries out there than to actually write or copy your own routines. The code in these libraries is relatively stable and has been thoroughly vetted. Why re-invent the wheel and re-implement code -- possibly incorrectly -- which is already available? The minor one is that this book does have a somewhat idiosyncratic c++ style (for example, exclusive use of "struct" without access specifiers, instead of the class keyword). To their credit, the authors themselves make these points early in the book.

The real strength of Numerical Recipes is in its exposition of how these numerical algorithms work. If all you need is an ODE integrator, this is probably not the best resource. However if you would like your code to be less of a "black box," and want to actually understand how different algorithms work, this is a peerless reference book. I am not aware of any other work which so thoroughly describes the internals of such a breadth of numerical algorithms.

I never bother to use the code from this book, but I've been using the second edition for years as a very solid mathematical and numerical reference. For me, this book is a critical reference and usually the first place I turn to when I need additional insight into methods I need but am not sufficiently familiar with.

I have owned copies of the first two editions of this book and I was impressed with the updates to this third edition. There are several new topics and the existing areas have been updated. I was able to use some of the statistical code in a production piece of software two days after receiving this new version. The website associated with the book also has a nice feature that figures out the header dependencies for you.