FEVAL error in implementing Newton's method.

This is M-File 1

function sol=newtonian(fn,dfn,x0,tol)
tol=0.000001;

old=x0+1; 
while abs (x0 - old) > tol; old=x0; 
  x0=old - feval(fn, old)/feval(dfn, old);

end; 

sol=x0;
end

M-File 2

function y=fn(x0,y)

    x0=2;
    y=(x0).^2-exp(x0)+2;
end

M-File 3

function dy=dfn(x0)
  x0=-2;
  dy=2*(x0)-exp(x0);
end

But I keep getting this ERROR:

??? Error using ==> feval
Argument must contain a string or function_handle.

Error in ==> newtonian at 5
x0=old - feval(fn, old)/feval(dfn, old);

What does this error mean?

[Paulo Silva]

??? Error using ==> feval

Argument must contain a string or function_handle.

means that you are not providing the arguments that the function is expecting, maybe feval(@fn, old) or feval('fn', old) ?

Answer 1

[Paulo Silva]

Here's one awesome resource you should view, it helped me a lot last year.

Answer 2

[Sean de Wolski]

 fn(old) 
 dfn(old)

Are what you want. Not feval.

[Ashley Dunn]

ok now it's saying my FN file is messed up ....like the

??? Input argument "x0" is undefined.

Error in ==> fn at 2

y=(x0).^2-exp(x0)+2;

[Sean de Wolski]

function y = fn(x0)

y=(x0).^2-exp(x0)+2;

end

or

fn = @(x0)x0.^2-exp(x0)+2;

[Paulo Silva]

There's also the inline function that works great

fn=inline('x0.^2-exp(x0)+2');

[Ashley Dunn]

Ok redid using what you said

>> x0=-2

x0 =

-2

>> tol=.01

tol =

0.0100

>> sol=newtonian(fn,dfn,x0,tol)

??? Error using ==> dfn

Too many output arguments.

And here are my modified m-files

:

function dfn(x0)

dfn=2*(x0)-exp(x0);

end

function fn(x0)

fn = @(x0)x0.^2-exp(x0)+2;

end

function sol=newtonian(fn,dfn,x0,tol)

old=x0+1;

while abs (x0 - old) > tol; old=x0;

x0=old - (fn(old))/(dfn(old));

end; sol=x0;

end

[Ashley Dunn]

what is inline

[Ashley Dunn]

I did inline for both dfn and fn

>> sol=newtonian(fn,dfn,x0,tol)

??? Error using ==> dfn

Too many output arguments.

[Paulo Silva]

doc inline

[Ashley Dunn]

what does doc inline mean?

[Paulo Silva]

write doc inline in your matlab command line and see, same goes to other functions. You can also do:

help inline

Answer 3

[Sean de Wolski]

Ashley, It would be worth your time to review the syntax and methodology covered in these two documents:

 doc function
 doc function_handle

[Ashley Dunn]

what is this??

Answer 4

[Matt Tearle]

The accepted answer from Paulo looks like a great resource for the theory behind gradient methods, but I'm not sure if that's actually solved your MATLAB issues or not, so...

The issue is that you need to give the names of two functions to newtonian (the first two arguments, fn and dfn). Values like x0 and tol are easy to pass to a function as either literal values or variables that hold numeric values. For example

sqrt(pi)

or

x = pi;
sqrt(x)

But how do you pass a reference to (or name of) a function as an input to another function? The answer in MATLAB is to use a function handle. If you have a function file (like, say, fn.m) you can create a function handle to it using the @ symbol: @fn

Then, inside newtonian the variable fn will be a function handle to the function file fn.m. You can then use it as if it was a direct use of the fn function, as in Sean's answer: fn(old).

(Sorry, I realize it's a bit confusing, with the same name being used for everything.)