Customer Reviews

Numerical Recipes in Fortran: The Art of Scientific Computing

5 star:		(6)
4 star:		(4)
3 star:		(0)
2 star:		(0)
1 star:		(0)

 

 

I first bought this text in 1994 while doing scientific programming for graduate school work. A fellow graduate student had suggested I use an undocumented routine that (I later discovered) came from Numerical Recipes (NR). I was impressed enough with NR's presentation of ideas that I also bought the example book ISBN 0521437210 (which I've hardly cracked since) and a diskette of source code (which cost as much as the book but worth it). I was able to do a lot of basic research quickly with NR code, and I still occasionally use NR's routines.

The authors have certainly done a good job assimilating a lot of material. Since other reviewers have done well to highlight the importance and utility of this landmark book, there is no need to repeat those sentiments here. However, to this title's detriment, the authors consider their book to be a proprietary library of source code more valuable than the explanatory text discussing it (one can in fact download the text on-line though it's hardly worth the hassle). This perception is ironic since the authors confess that "the lineage of many programs in common circulation is often unclear" (p.xviii), and many details of presentation, ideas, and algorithms are clearly "borrowed" from other excellent (some now out-of-print) numerical methods books or journals.

I often wondered why NR routines occasionally adopted bizarre and/or obviously inefficient programming structures - over time I decided that this was probably done to make these algorithms appear as so not to clearly violate other published material. As a student, NR's legal disclaimers regarding derivative works (p.xvi) never bothered me and I was willing to overlook the sometimes unpolished source code insofar as it functioned properly. However, as a professional I now find the lack of fair-use provisions on the uncompiled source way too restrictive to rely on these routines in good conscience (I have to buy another textbook or license for every soft copy or machine upon which the source code resides!). I suspect this policy ultimately hurts NR's textbook sales: it would be nice to able to use and pass along the source code between professional colleagues without restriction because most would certainly buy (if they don't already own) the textbook to understand what the source does (just as I did). Source code used in scientific programming is practically worthless without proper documentation, and there's no better documentation than a full length textbook!

I have since expanded my numerical methods library to other references supporting true public-domain codes. With an expanded basis of comparison, I regret to say that I am becoming less and less impressed with NR's implementations and explanations. I am finding many of NR's algorithms to be inefficient or unnecessarily approximate, and - on rare occasion - buggy. There have been quite a few bugs uncovered over the years, and the NR web site has done a good job of keeping track of them (although I know of at least one bug uncorrected by NR to this day).

This book is excellent for students wanting a good reference for quick and dirty types of analyses or scientific computing. Professional programmers, scientists, engineers, specialists or analysts performing software development for laboratory or scientific research would be well advised to reference this title, but ultimately they will likely need to rely other resources if they require efficient and/or unrestricted (public-domain) source codes for their work.

(P.S. - A reviewer elsewhere noted that the "quality of the binding was terrible" and I've also found this to be the case. My hardcover is literally had to be taped on after a few years of use.)

I converted the authors' multigrid program on p. 870 to Visual Basic and ran it in EXCEL. Comparison with exact solutions for special cases of Poisson's equation showed it to be accurate to micro percents in less time than it takes to write the spreadsheet (a second or so). The explanation of the program in the text is clearer than specialist discussions like Wessling (An Introduction to Multigrid Methods), but doesn't contain theoretical analysis of convergence rates. Personally, I find it more persuasive to plot the errors vs number of cycles, number of smoothings etc. than to read theoretical analysis.

The authors also compare various methods and give practical advice about which methods to use. Specialist texts on numerical methods aren't much good in this regard, being hung up on methods where it is easy to prove convergence rates.

Bottom line: Good code, good qualitative discussion, good comparisons of methods.

This book provides a brief description of a series of numerical methods, and their respective fortran codes. It is a toolbox for anyone interested in apply the numerical method rather than work on it.

Most of the routines are reliable, even though they are not state of the art. It is a good aquisition. Naturally, the disk is a must if you do not want type (or mistype) the source codes.

I purchased the "Numerical Recipes The art of Scientific Computing" to learn more about implementing Fourier transforms into my programming applications. I hit the jackpot with this series of books. The companion workbook "Numerical recipes, Routines and Examples in BASIC" provides examples of BASIC code in a printed format. A BASIC (IBM) diskette will save a lot of time typing. Ordering information is provided in the back of the book for obtaining the accessories. The "Numerical Recipes The art of Scientific Computing" series is also available in Pascal and C programming language flavors. This book series is the mother lode for paste-in numerical data reduction and processing routines.

This volume, and its companions for other programming languages, is an absolute classic. The authors strike the right balance between cookbook solutions and theory, so that most of us get just enough background to choose the right algorithm but not so much to get drowned in theory. This edition is the first devoted only to Fortran, but is the second edition published by the authors. It includes a number of additions and corrections, many of which appeared in Computers in Physics (now the journal Computing in Science and Engineering published jointly by the IEEE and the APS). My only criticism is, where were these books twenty years ago when I needed them? I would recommend these books to anyone involved in the application of numerical methods. They are tremendous time savers.

I never bothered with the discs, as most of the routines are fairly short and not a problem to type in, but I recommend the companion example books to help get the routines running.

If you ever had to program a complicated numerical algorithm, such as SVD decomposition, Bessel functions, eigensystems or Fourier transform, you will know how useful this book is. All those problems, and many others, are presented, the theory is explained and the full code of a routine, which solves it, is given. This version brings the codes in FORTRAN 77, but there are versions for Pascal, C++ and Basic. If you need any routine, you just have to "cut and paste" it from the book into your program.

I've had this book for a long time and, unlike some other's copies, it's still intact. The exposition is so well done that I occasionally read passages just for
entertainment (a rarity for me). I am now implementing some routines from the
f90 "second edition" and expect to get faster results (but can't confirm that yet).
On the irritating "licencing" issue: Some years ago my company attempted to "do the
right thing" and obtain a site license for some of the f77 routines. We never got
a response back! I have a hunch that some lawyers wrote the licensing text but that
maintaining a licencing infrastructure was too much trouble Press et al to bother with. I've encountered a similar non-response when trying to license some songs
from song writers for a limited release DVD.

First, I want to weigh in on the general controversy over Press et al.'s treatment of the code as proprietary. This is a joke and I think was basically intended as such. A lot of the odd job programming world still works on the honor system, and it works, because enough of us respect what is done right and we respect other people's labor. But if you've already plunked down your $60 for the book, and you have a program that in some sense you are paid for (not that you're Adobe or someone big), they expect you to be crawling to them for permission to use a single routine. And yet these routines make use of other routines that I really doubt they wrote. I see the same basic code floating around a number of places. And for some things, it's so straightforward it seems crazy to say that it is "their" routine.

At the same time, the code is clear, well explained with examples, and--most important--easy to modify. Given a choice between two versions, one which was elegant but a bit confusing and one that was straightforward, they did the second.

Given that there's a new edition, and that we are 30 years past FORTRAN 77, you might wonder whether this book should still be in circulation. But there are a lot of industrial machines that are still running programs written in 77, say for process control, and it's nice to be able to alter them as opposed to starting from scratch. For example, you might find that a minimization routine actually can help for a program running the control of a valve in a mixing vat. Some of these routines are slow, but my attitude is, that just gives you more time to drink your coffee and look around, though of course, you might not like what you see. [12]

This is the single best book that I have found for teaching numerical methods in science and engineering to upper division undergraduates and graduate students. Students often comment that this should be the selected text even in the programming course because it provides both an overview of the methods and examples that demonstrate the application. The discussions are excellent and the Fortran 77 programs easy to follow even if one is more familiar with C or C++. You should not purchase the Fortran 90 version of this book without getting this book as well because the Fortran 90 book does not contain the excellent discussion of the methods and procedures. Rather it references this book for discussion and simply provides the F90 versions of the routines.

This book contains hundreds of "canned codes" in the FORTRAN language. The book provides several variations of many popular numerical techniques and provides the most stream line (comp. time) codes available. Most codes allow for optimization to be build in, such as an RK4 (4th Order Runge-Kutta) with variable steps sizes. Great if you don't want to write your own code for a subroutine, or it you just don't know the method well enough to write it yourself. The book also provides some basic explaination of the techniques and codes with is very helpful so that the code is less of a black box, although its not that detailed.

There is also a CD available that has the codes already written and ready to go. I prefer to type it in on my own, or just make my own because it gives a better udnerstanding of what the code is doing. The biggest turn-off for me is that some codes have subroutines upon subroutines which can make things a mess.

All around a useful tool for programmers, researchers, and students.