The value returned by NMaximize increases as the domain decreases?

Questions
,
1

Definition of the function g:

f2[x_] := (x*(FresnelC[x] + FresnelS[x]) + (Cos[Pi*x^2/2] - 
       Sin[Pi*x^2/2])/Pi)/2

g[x1_, x2_, y1_, y2_] := 
 2*(f2[x2 - y1] - f2[x1 - y1] - f2[x2 - y2] + f2[x1 - y2])/(x2 - x1 + 
     y2 - y1)

First maximization with a suitable domain:

NMaximize[{g[x1, x2, y1, y2],  0 < x1 < x2 < 7 && -7 < y1 < y2 < 0}, {x1, x2, y1, y2}]


Out[33]= {0.00937548, {x1 -> 1.60404, x2 -> 5.27857, y1 -> -2.89746, 
  y2 -> -0.0330524}}

Second maximization with a smaller domain: why the maximum is larger???

In[34]:= NMaximize[{g[x1, x2, y1, y2],  0 < x1 < x2 < 5 && -5 < y1 < y2 < 0}, {x1, x2, y1, y2}]


Out[34]= {0.0260921, {x1 -> 1.63214, x2 -> 3.49276, y1 -> -1.26015, 
  y2 -> -0.0177181}}
2

NMaximize likely got trapped in a local max. Might try different Method option settings to see if it can be made to do better. Here is one such attempt. Note that it pushes some variables to the boundary of the allowed region.

In[9]:= NMaximize[{g[x1, x2, y1, y2], 
  0 < x1 < x2 < 7 && -7 < y1 < y2 < 0}, {x1, x2, y1, y2}, 
 Method -> "DifferentialEvolution"]

Out[9]= {0.401025, {x1 -> 0., x2 -> 0.707106, y1 -> -0.707108, 
  y2 -> 0.}}
3

Thank you! I am new to Mathematica, but I see that the situation is more complicated than I thought.