0
$\begingroup$

I've encountered a very strange problem. I'm trying to find a minimum of a real part (or absolute value, or really any possible real-valued expression) of some complex function, which was previously interpolated from numerical data. Let's see here for the minimal "working" example:

func = Interpolation[Table[{x, Exp[I x]}, {x, -5, 5, 0.01}]];
NMaxValue[Re[func[x]], x]

The "result" of this is:

NMaxValue::cvmit: Failed to converge to the requested accuracy or precision within 100 iterations.
-Experimental`NumericalFunction[{x}, -Re[
     InterpolatingFunction[{{-5., 5.}}, <>][x]], 
   "-NumericalFunctionData-"][{-6.42031}]

Can someone point me in the right direction and hint on what the problem might be? It works when I do

func = Interpolation[Table[{x, Cos[x]}, {x, -5, 5, 0.01}]];
NMaxValue[func[x], x]

but my functions are usually complex.

Behavior is the same with NMaximize. What am i missing here? What is wrong with Re? What exactly is Experimental``NumericalFunction? Thanks.

$\endgroup$
1
$\begingroup$

(too long for a comment)

What you saw was NMaximize[] (and thus NMaxValue[]) trying to create a function that is optimized for numerical evaluation, since it is expecting the objective function to be evaluated a lot of times (especially since NMaximize[] uses stochastic methods).

The cure (see this as well), as is usual for thwarting Mathematica's attempts to be smart with numerical function evaluations, is

func = Interpolation[N@Table[{x, Exp[I x]}, {x, -5, 5, 1/100}]];
ff[x_?NumericQ] := Re[func[x]]

NMaxValue[{ff[x], -5 < x < 5}, x]
   1.

Note that I had also taken the liberty to constrain the independent variable; since you using an interpolating function anyway, the constraint ensures no extrapolation gets done.

$\endgroup$
  • $\begingroup$ Well, as far as I'm concerned, this is the (or at least 'an') answer! Not sure who will get the rep for it tho, upvoting anyway. I will wait with accepting this as not to discourage anyone in case more solutions would come up. $\endgroup$ – user16320 Mar 12 '18 at 12:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.