3
$\begingroup$

I am trying to use NMaximize to find the maximum value of a variable that satisfies the given constraints. Since the constraints aren't straightforward, I am using the function.

I can see the constraints are such that the value is bounded but I get the below warning messages:

NMaximize::cvmit: Failed to converge to the requested accuracy or precision within 100000 iterations.

NMaximize::cvdiv: Failed to converge to a solution. The function may be unbounded.

The constraint and the way I am using the function is as below:

    constraint = (x | y) \[Element] 
  Integers && ((x == 0 && 1. <= y <= 12720.) || (1. <= x <= 10712. && 
     0 <= y < 
      2.08565*10^-36 (3.04959*10^39 + 2.24751*10^34 x) + 
       2.8484*10^-43 Sqrt[
        4.98614*10^92 + 4.65469*10^88 x - 
         3.63201*10^84 x^2]) || (10713. <= x <= 19762. && 
     2.08565*10^-36 (3.04959*10^39 + 2.24751*10^34 x) - 
       2.8484*10^-43 Sqrt[
        4.98614*10^92 + 4.65469*10^88 x - 3.63201*10^84 x^2] < y < 
      2.08565*10^-36 (3.04959*10^39 + 2.24751*10^34 x) + 
       2.8484*10^-43 Sqrt[
        4.98614*10^92 + 4.65469*10^88 x - 3.63201*10^84 x^2]))


maxX =  
 NMaximize[{x, constraint}, {x, y}, MaxIterations -> 100000]

I have increased the MaxIterations from 100 to 100000 but it doesn't seem to converge. I am not sure if increasing the MaxIterations is the solution. Can you please guide me with this?

Improve this question
$\endgroup$
6
  • $\begingroup$ Could try maximizing over individual regions of the piecewise set-up. But the machine precision values will make validation of inequalities kind of iffy. $\endgroup$
    – Daniel Lichtblau
    Commented Apr 5, 2019 at 17:44
  • 1
    $\begingroup$ I'm not seeing what $y$ has to do with this. Wouldn't the maximum value of $x$ be 19762? constraint /. x -> 19762 results in y \[Element] Integers && 7229.16 < y < 7344.29 and constraint /. x -> 19763 results in False. $\endgroup$
    – JimB
    Commented Apr 5, 2019 at 17:51
  • $\begingroup$ @JimB, I think for x, y isn't needed. Thanks for pointing this out. But if I am trying to maximize y, I need to maximize over both the variables since y is an expression of x, right? $\endgroup$
    – gaganso
    Commented Apr 5, 2019 at 18:06
  • $\begingroup$ Yes, if that's what you want. The general solution appears to be $x = 19762$ and $7230\leq y \leq 7344$. So to maximize $y$ you'd choose $7344$. $\endgroup$
    – JimB
    Commented Apr 5, 2019 at 18:49
  • 1
    $\begingroup$ OK. I was assuming that you were conditioning on the maximum value of $x$. $\endgroup$
    – JimB
    Commented Apr 5, 2019 at 18:57

2 Answers 2

Reset to default
5
$\begingroup$

Rationalize the constraint:

constraint2 = ((x == 0 && 1. <= y <= 12720.) || (1. <= x <= 10712. && 
        0 <= y < 2.08565*10^-36 (3.04959*10^39 + 2.24751*10^34 x) + 
          2.8484*10^-1 Sqrt[
            4.98614*10^8 + 4.65469*10^4 x - 3.63201 x^2]) || (10713. <= x <= 
         19762. && 
        2.08565*10^-36 (3.04959*10^39 + 2.24751*10^34 x) - 
          2.8484*10^-1 Sqrt[4.98614*10^8 + 4.65469*10^4 x - 3.63201 x^2] < y <
          2.08565*10^-36 (3.04959*10^39 + 2.24751*10^34 x) + 
          2.8484*10^-1 Sqrt[4.98614*10^8 + 4.65469*10^4 x - 3.63201 x^2])) // 
    Rationalize[#, 0] & // Simplify;

With the Rationalized constraint you can use Maximize:

maxX = Maximize[{x, constraint2}, {x, y}]

(* {19762, {x -> 19762, y -> 7287}} *)

constraint2 /. maxX[[2]]

(* True *)

EDIT: To find maximum y

(maxY = Maximize[{y, constraint2}, {x, y}]) // N

enter image description here

To plot the region defined by the constraint:

reg = ImplicitRegion[constraint2, {x, y}];

Region[reg,
 Frame -> True,
 FrameLabel -> (Style[#, 12, Bold] & /@ {x, y}),
 Epilog -> {Red,
   AbsolutePointSize[3],
   Point[{x, y} /. maxX[[2]]],
   Point[{x, y} /. maxY[[2]]]}]

enter image description here

Improve this answer
$\endgroup$
3
$\begingroup$

You have numbers spread a wide range of magnitudes for no good reason. This range is probably too wide for machine precision arithmetic. Also telling NMinimize explicitly that this an integer optimization problem seems to help. Try this:

constraint2 = ((x == 0 && 1. <= y <= 12720.) || (1. <= x <= 10712. && 
      0 <= y < 
       2.08565*10^-36 (3.04959*10^39 + 2.24751*10^34 x) + 
        2.8484*10^-1 Sqrt[
          4.98614*10^8 + 4.65469*10^4 x - 3.63201 x^2]) || (10713. <= 
       x <= 19762. && 
      2.08565*10^-36 (3.04959*10^39 + 2.24751*10^34 x) - 
        2.8484*10^-1 Sqrt[
          4.98614*10^8 + 4.65469*10^4 x - 3.63201 x^2] < y < 
       2.08565*10^-36 (3.04959*10^39 + 2.24751*10^34 x) + 
        2.8484*10^-1 Sqrt[
          4.98614*10^8 + 4.65469*10^4 x - 3.63201 x^2])) // Expand

maxX = NMaximize[{x, constraint2}, {x, y}, Integers, 
  MaxIterations -> 10000]

{19762., {x -> 19762, y -> 7311}}

And with your definition of constraint:

constraint /. maxX[[2]]

True

Improve this answer
$\endgroup$
3
  • $\begingroup$ But constraint /. x -> 19762 /. y -> 8647 results in False? $\endgroup$
    – JimB
    Commented Apr 5, 2019 at 17:57
  • $\begingroup$ @JimB D'oh. Yeah, I did the simplification wrong. -.- Thanks for pointing that out. $\endgroup$
    – Henrik Schumacher
    Commented Apr 5, 2019 at 18:02
  • $\begingroup$ @HenrikSchumacher, thank you for this. This works for x but when I try to find the maximum y similarly, I still get the same message - NMaximize[{y, res}, {x, y}, Integers, MaxIterations -> 100000]. Output: NMaximize::cvdiv: Failed to converge to a solution. The function may be unbounded. $\endgroup$
    – gaganso
    Commented Apr 5, 2019 at 18:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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