0

I have a following issue. I am trying to create a function handle, which is a vector. In particular, I have something like this

    EQ0 = @(W) m1.^2*(exp(W))-m2.^2

where m1 and m2 are the vectors of the same dimension. So, for each m1(i) and m2(i) I want to have a handle W(i). I need it in order to find those W(i)'s in the next step using fsolve in something looking like this

    n=size(m1)        
    x0 = zeros(n);
    Wbar = fsolve(EQ0,x0)

I have tried using arrayfun, but received a following error

   EQ0 = arrayfun( @(W) m1.^2*(exp(W))-m2.^2, m1=m1e, m2=m2e)
   Error: The expression to the left of the equals sign is not a valid target for an assignment.

Another attempt in using arrayfun resulted in this (here I just used m1 and m2 vectors directly, not as an inputs like in previous case)

    EQ0 = arrayfun( @(W) m1.^2*(exp(W))-m2.^2,:)
    Undefined variable arrayfun.

I am clearly missing something. I have looked on some feeds on arrayfun but it looks like my problem is somewhat different.

Any advice is appreciated.

Natalia K
  • 3
  • 3
  • isnt the function handle rather EQ0 than W? if looping an option here, so that you would have EQ0(1:10) and x0(1:10) and the loop through the 10 different? – Finn Jul 06 '16 at 12:39
  • This is **not** Python. `m1=m1e` and `m2=m2e` is not correct syntax. – rayryeng Jul 06 '16 at 20:10
  • @Finn My sole objective here is to eventually solve a system of independent non-linear equations for W. I define a function f(W, m1, m2, ...) where W would be a matrix input (n x m), and then I want to assign a handle to operate this matrix W, in order to be able to use fsolve to find those Ws (think of fixed-point problem for each entry of this (n x m) matrix) – Natalia K Jul 07 '16 at 10:27

1 Answers1

0

So if I understood you right you want to have for each m1(i) or m2(i) a seperate function handle EQ0(i) which can operate with an vector W in the following way EQ0(i) = @(W) m1(i)^2*(exp(W))-m2(i)^2. Is this correct? If so you can create a cell-array of the same dimension as m1 with a function handle in each dimension:

EQ0 = cell(size(m1));
for ii = 1:numel(m1)
   EQ0(ii) = {@(W) m1(ii)^2*exp(W)-m2(ii)^2};
end

EDIT: Another option could be:

EQ0 = @(W)arrayfun(@(Wel,m1el,m2el)m1el^2*exp(Wel)-m2el^2,W,m1,m2);
fsolve(EQ0, W_Values)

Here m1, m2 should be defined beforehand. Otherwise you have to add them as arguments of the first anonymous function. So by calling arrayfun(@(Wel, m1el, m2el)..., W, m1, m2) you do your elementwise calculations for the entries in W, m1, m2 defined by the anonymous function handle you have passed in the first argument of arrayfun. But since you want to define your W differently each time, you make an anonymous function of that arrayfun command, which takes W as an argument.

v.tralala
  • 1,444
  • 3
  • 18
  • 39
  • This seems to be a solution to the problem. If there is a way to do it without loops, I would like to know how for the future (if it means being faster). – Natalia K Jul 07 '16 at 12:44
  • Just a comment: even though the thing above works, since doing above in a loop and then applying fsolve in a loop (my matrices have over 100,000 individual entries) is quite inefficient. Still waiting for a code to finish running.... – Natalia K Jul 07 '16 at 17:23
  • Ah ok I think what you are looking for is something like the thing I've edited? – v.tralala Jul 10 '16 at 12:04
  • Yes. The edited version is much more useful for large sets of data since it avoids loops. – Natalia K Jul 11 '16 at 13:19
  • Careful. `arrayfun` is still technically a loop. – rayryeng Jul 16 '16 at 13:57
  • @rayryeng So you mean arrayfun is not paralellized? – v.tralala Jul 25 '16 at 22:46
  • Exactly... unless you use gpuArrays – rayryeng Jul 25 '16 at 23:05
  • Ah I found some interesting post about this: [link](http://stackoverflow.com/questions/12522888/arrayfun-can-be-significantly-slower-than-an-explicit-loop-in-matlab-why) Thanks for pointing this out! – v.tralala Jul 25 '16 at 23:10