That's what I get for all that Scheming

So I know that I have sang the praises of Scheme in several posts now, but let me get to some of the downsides:

First of all, it's insane. No control structures. No local variables. THE MADNESS!!!

But more importantly for me, because it uses exact numbers (ie fractions), the numbers can quickly become to large for the interpreter to handle within reasonable amounts of time. In many cases having exact numbers is ideal for math, but when your using mathematical estimates based on numerical analysis that is iterated many, many times, well, things get ugly.

So I'm putting out this question: Does anyone know of a way to turn off exact numbers in Scheme. I realize that you can just use the function (inexact->exact num), but that only turns it off for one instance, what I'm looking for is a more general purpose off switch.

Because of this difficulty, for some of the more advanced numerical methods I'm handling I'm switching to OpenOffice Spreadsheet. It's not a perfect system, but if properly used the spreadsheet can be a mighty powerful tool of data manipulation.

And I shall also see if I can design functions that get around the inexact/exact difficulty even if it is with the inexact->exact function, because I dream, I dream of the day when once more I can Scheme.

Time keeps on Schemin'

So one thing I'm a little bit worried about with this blog is the temptation to turn it into a semi-computer science blog. I want to keep things well rooted in math here, because math is awesome. I mean computer science is cool and all, but math is awesome, and also I think there are more computer science blogs out there than math blogs.

Still, I'm dealing with a lot of numerical analysis stuff now, the line between math and computer science is a little bit blurry. Still, because this blog is mine, I'll walk that line.

And so in the interest of math, I will now reveal to you some really cool Scheme code (just to refresh your memories, I'm using Dr. Scheme, a free Scheme interpreter, and I've been consulting The Scheme Programming Language by R. Kent Dybvig for reference) I wrote to deal with numerical analysis problems. Many people have said that Scheme should not be used for numerical analysis hw, and they are right, but that's why it's ultra-cool when you get it right.

So here's some functions:

General method applier for numerical methods requiring iteration (for generating a list of the value at several different iterations, largely for hw purposes but also so you can look at how it is converging).

(define (applymthd method fun initx tol)
(
map (lambda (iter) (method fun initx tol iter)) '(1 5 10 20 50 100 1000)
)
)

So this obviously is a pretty rough function, but I can tweak it into shape and if I do, I'll pass it along.

Here's a cheeky little function for forward distance:

(define (fordist fun n)
(
(- (fun (+ n 1)) (fun n))
)
)

Here's a less cheeky function for Aiken's terms given a function:

(define (Aitkens fun n)

(- (fun n) (/ (expt (- (fun (+ n 1)) (fun n)) 2) (+ (fun (+ n 2)) (* -2 (fun (+ n 1))) (fun n))))

)

But moving on to more sizable methods, here's an implementation of Steffensen's method:

(define (Steffensens fun initx tol iter)
(if (< iter 1)
initx
(
if

(< (abs (/ (expt (- (fun initx) initx) 2) (+ (fun (fun initx)) (* -2 (fun initx)) initx))) tol)

(- initx (/ (expt (- (fun initx) initx) 2) (+ (fun (fun initx)) (* -2 (fun initx)) initx)))

(Steffensens fun
(- initx (/ (expt (- (fun initx) initx) 2) (+ (fun (fun initx)) (* -2 (fun initx)) initx)))
tol (- iter 1))
)
)
)

More stuff: An implementation of Fixed Point Iteration:

(define (fixedpnt fun initx tol iter)
(if (< iter 1)
initx
(if (< (abs (- (fun initx) initx)) tol)
(fun initx)
(fixedpnt fun (fun initx) tol (- iter 1))
)
)
)

So that's that for now. And I'd say that's a good chunk of stuff, so that's some Scheme, but really, it's all about how to deal with numerical analysis in a quick and painless fashion made slightly more insane by using Scheme instead of any number of more saner tools.

Why plan when you can scheme?

Let me make a pitch for programming in scheme for math purposes.

Why is Scheme good for programming numerical methods and other math brick and brack?

Because it is so gloriously insane!

Scheme is a language without local variables, without control structures, without global constants, it uses reverse-Polish notation (those crazy Poles!) run usually by an interpreter... basically to the layman, let me say that it makes no sense.

But the advantage to Scheme is it fits very well with the idea of constructing logical algorithms and basing your math around that. Scheme strips algorithms down to their basic components, the procedures, and doesn't allow you to go any further than that. There's a ruthlessness to this restriction and enough power in this limited space that you end up being able to do a lot of math stuff simply and in a manner which illuminates the heart of the algorithm.

But it remains insane.

That said here's a good Scheme implementation:

Dr. Scheme

And here's a good guide to the language (note the implementation and this guide are not related, so some aspects of the guide may not work with this implementation):

The Scheme Programming Language by Kent Dybvig

Now I don't want this math blog to go all computer sciency, but it must be admitted that comp. sci. has some math overlap, especially when it comes to doing programming to make math easier. So look out in the future for some bursts of code for math-based functions, most likely in Scheme.

Keep on Scheming!