Browse Books

Reviewer: Fernando Berzal

The "Numerical Recipes" series on numerical algorithms for scientific computing is on the short list of books that every computing professional knows. Experienced programmers often refer to these books as the authoritative source on numerical computing. They still live up to their reputation, even in today's Web-based world, where you can Google any algorithm to get its open-source implementation, in almost any programming language. However, since all Web references are not equally reliable, these books have maintained their privileged position, despite the years that elapsed between the editions. Apart from the printed book, the authors also offer an electronic version of their book by subscription, implicitly acknowledging the suitability of the Web as the best medium to provide an up-to-date algorithm collection. Moreover, without the page limitation of a single volume (now with more than 1,200 pages), the electronic version can grow to become the definitive reference work on numerical algorithms. What is behind its excellent reputation as a reference book__?__ It is certainly not the code provided (more on that later). This book is invaluable because of the concise and clear explanations that provide the minimum amount of information needed to understand the rationale behind each numerical technique. It is also priceless because of the critical discussions of alternative methods, which provide useful advice on when (and when not) to use each one. Unfortunately, the provided C++ implementation of the different algorithms, albeit useful for a quick-and-dirty solution, can make you cringe, especially if you are an object-oriented zealot. All of the source code is provided in many interdependent header files. Sometimes, its use of structs and class templates for grouping functions seems irrational, as does the use of inheritance as a "grab-bag" for passing sets of methods to an object. Additional details, such as the ample use of abbreviated identifiers instead of more descriptive ones, make me think that while the authors have undisputed experience in scientific computing, they have not grasped the main ideas behind object-oriented programming. Their C++ implementations read like old C/Fortran code, and it is not clear to me why they decided to use C++, given their old-school implementation style. Despite the aforementioned implementation flaws, we should be grateful that the authors have updated the aging second edition, and have now included new sections on topics such as mixture models, hidden Markov modeling, phylogenetic trees, support vector machines, and computational geometry, apart from the many updates and improvements of topics already covered 15 years ago. This is still the book to consult for any number of tasks, including: algorithmic techniques for solving algebraic or differential equations; interpolating data, and evaluating or integrating functions; solving optimization problems; applying signal processing transformations such as fast Fourier transforms (FFTs) and wavelets; statistically modeling data; or generating pseudo-random numbers (beyond the limitations of the linear congruential generators). Rather than complete recipes, the book provides all the ingredients you may need to solve numerical computation problems. In the authors' own words, the book "is intended as a cookbook for cooks, not a restaurant menu for diners." Any interested reader will find clear descriptions of the fundamentals behind the techniques described in each section, along with suitable bibliographic references that will satiate your need to know the details. This beautifully printed book, which uses color for headings and code listings, deserves its place on the shelf of any self-respecting scientific programmer. Online Computing Reviews Service