setting NaNs to zero in an anonymous function inline

Question

I implemented a piecewise function using an anonymous function that can also be evaluated in vectors or matrices:

 f = @(x) exp(((x-.25).*(x-.75)).^-1) .* (.25 < x & x < .75);

The problem is when evaluating it at exactly 0.25 or 0.75 we try to evaluate Inf * 0 which results in NaN. I would like to set these NaN values to zero. I am aware that this can easily be done using a function defined in a file, but I'm wondering whether there is a solution that would let you do this inline. (Possible defining other anonymous helper functions.)

Answer 1

To get rid of the NaN's, you can exploit the fact that some functions like max ignore NaN's:

>> max(NaN, 0)
ans =
     0

which also works vectorized:

>> max([1 2 NaN 4], 0)
ans =
     1     2     0     4

---

The mathematical function you are working with is non-negative. So if y is an array resulting from the anonymous function as defined in your question, max(y, NaN) will leave numeric entries of y untouched, and will transform NaN into 0.

In conclusion, modify the anonymous function to include an outer max(..., 0):

f = @(x) max(exp(((x-.25).*(x-.75)).^-1) .* (.25 < x & x < .75), 0);

This gives, for example,

>> f([.2 .25 .4 .5 .6 .75 .8])
ans =
   1.0e-06 *
         0         0    0.0053    0.1125    0.0053         0         0