Register to reply

Computing derivatives/integrals of user-defined functions

by clope023
Tags: computing, functions, userdefined
Share this thread:
clope023
#1
Mar5-09, 12:58 PM
P: 618
does anyone have ideas or easy to read versions of code for this? I was looking for ways to do it using the limit definition of a derivative of a function and I've found a few applets but nothing that is that clear; any ideas are appreciated.
Phys.Org News Partner Science news on Phys.org
Flapping baby birds give clues to origin of flight
Prions can trigger 'stuck' wine fermentations, researchers find
Socially-assistive robots help kids with autism learn by providing personalized prompts
csprof2000
#2
Mar5-09, 02:12 PM
P: 287
In general, you can adopt an approximate numerical approach, or an exact symbolic approach.

In the numerical approach, all you need to do is have some way of evaluating a user-defined function for a given value of x. You could write this by yourself (you would have to write a parser that would recognize well-formatted expressions, like a calculator), or you could use already existing code (you should be able to find lots of this kind of thing). So I enter

f(x) = 2*x + (exp(-3*x+2)) and I want f '(3)...

You approximate the derivative using the secant method. Basically,

f '(3) ~ f(3 + d) - f(3 - d) / 2d

You will want to play around with the value of d, but that's about all there is to it.


The symbolic approach, on the other hand, will change everything to tokens, and recursively differentiate the tokens. The base cases will consist of polynomial expressions, sines and cosines, exponentials, etc. Then just build in to the interpreter the properties of differentiation - you know, like the product, chain, quotient, power, etc. rules - and then you can change the user function into a new function.

Then you can just evaluate the derivative function at the desired point.

The answer to your question depends on which strategy you find more appealing.
clope023
#3
Mar5-09, 04:45 PM
P: 618
Quote Quote by csprof2000 View Post
In general, you can adopt an approximate numerical approach, or an exact symbolic approach.

In the numerical approach, all you need to do is have some way of evaluating a user-defined function for a given value of x. You could write this by yourself (you would have to write a parser that would recognize well-formatted expressions, like a calculator), or you could use already existing code (you should be able to find lots of this kind of thing). So I enter

f(x) = 2*x + (exp(-3*x+2)) and I want f '(3)...

You approximate the derivative using the secant method. Basically,

f '(3) ~ f(3 + d) - f(3 - d) / 2d

You will want to play around with the value of d, but that's about all there is to it.


The symbolic approach, on the other hand, will change everything to tokens, and recursively differentiate the tokens. The base cases will consist of polynomial expressions, sines and cosines, exponentials, etc. Then just build in to the interpreter the properties of differentiation - you know, like the product, chain, quotient, power, etc. rules - and then you can change the user function into a new function.

Then you can just evaluate the derivative function at the desired point.

The answer to your question depends on which strategy you find more appealing.
interesting, this is what I tried to write for a derivative function but it gave me a 0 so obviously it wasn't correct (in C)

int diff(int function, int h)
{
function = ((function+h)-function)%h;
return 0;
}

you'll have to forgive me but I'm not following how the symbolic approach is done

signerror
#4
Mar5-09, 07:45 PM
P: 223
Computing derivatives/integrals of user-defined functions

Quote Quote by clope023
int diff(int function, int h)
{
function = ((function+h)-function)%h;
return 0;
}
Several issues:

* That function returns 0.
* It has integer types, where you probably want to use a floating-point type like 'float' or 'double'.
* '%' is a modulus operator; '/' is division
* Your 'function' argument is of integer type ('int'), not function type. It is a number, not a function of a number.

The last issue is most significant, and problematic. You will, unfortunately, need to use C's function pointers to implement these kinds of things, functions which take functions as arguments. You will probably want to look this up in whatever C reference you have. For example, differentiation:

double diff(double (*f)(double), double x, double dx) {
  return ((*f)(x + dx) - (*f)(x)) / dx;
};
Or you can hack together a macro, which is somewhat simpler:

#define diff(f, x, dx) ((f((x)+(dx)) - f(x))/(dx))
Both approaches are difficult and dangerous (in the sense, it is easy to make errors and not be able to find them). The C language was not designed to support this type of programming. If you know any high-level languages, it becomes much easier. Python, for instance - very easy-to-use, easy-to-learn language:

def diff(f, x, dx):
   return (f(x+dx) - f(x)) * 1.0 / dx
(Note the lack of type annotations - Python will infer that f is a function, and x and dx are numbers. The 1.0 is there to coerce floating point division, as opposed to integer division with rounding.)

Symbolic differentiation is much harder. To implement it in C, you will need to figure out a way to represent the structure of a symbolic expression as a C data structure (probably a tree, made with pointers). Then you will need to write a function to differentiate such an expression, probably acting recursively - taking the expression, breaking it up, differentiating the parts, and combining them using (e.g.) the product rule.

However, it is easier if you use a symbolic programming language, like the educational language Scheme. For example, there is a worked example of symbolic differentiation in chapter 2 of the free book "Structure and Interpretation of Computer Programs". It's not particularly easy, but if you're motivated...

http://mitpress.mit.edu/sicp/


Register to reply

Related Discussions
Piecewise-defined functions Calculus 2
All of the functions that achieve defined Calculus & Beyond Homework 7
A matlab question(about user defined functions) Math & Science Software 0
Starting point not defined for contour integrals? Calculus 1
Well defined integrals Calculus & Beyond Homework 2